├── assets
    └── example_grey.png
├── src
    ├── interface
    │   ├── MethodImplementation.h
    │   └── AbstractBinarizer.h
    ├── utils
    │   ├── CudaUtil.h
    │   ├── RunConfigurationBuilder.h
    │   ├── ImageFileUtil.h
    │   ├── RunConfigurationBuilder.cpp
    │   ├── CudaUtil.cpp
    │   └── ImageFileUtil.cpp
    ├── model
    │   ├── ExecutionTimestamp.h
    │   ├── BinarizationResult.h
    │   ├── PngImage.h
    │   ├── RunConfiguration.h
    │   ├── PngImage.cpp
    │   ├── ExecutionTimestamp.cpp
    │   ├── BinarizationResult.cpp
    │   └── RunConfiguration.cpp
    ├── core
    │   ├── binarizers
    │   │   ├── SMCudaOtsuBinarizer.cuh
    │   │   ├── MonoCudaOtsuBinarizer.cuh
    │   │   ├── OtsuBinarizer.h
    │   │   ├── OtsuOpenMPBinarizer.h
    │   │   ├── CudaOtsuBinarizer.cuh
    │   │   ├── OtsuBinarizer.cpp
    │   │   ├── OtsuOpenMPBinarizer.cpp
    │   │   ├── SMCudaOtsuBinarizer.cu
    │   │   ├── MonoCudaOtsuBinarizer.cu
    │   │   └── CudaOtsuBinarizer.cu
    │   ├── AppRunner.h
    │   └── AppRunner.cpp
    ├── main.cpp
    └── libs
    │   └── lodepng.h
├── Makefile
├── LICENSE
└── README.md


/assets/example_grey.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/palucdev/CudaOtsu/HEAD/assets/example_grey.png


--------------------------------------------------------------------------------
/src/interface/MethodImplementation.h:
--------------------------------------------------------------------------------
1 | enum MethodImplementation : unsigned int
2 | {
3 | 	CPU,
4 | 	CPU_OpenMP,
5 | 	GPU,
6 | 	GPU_SharedMemory,
7 | 	GPU_MonoKernel,
8 | 	ALL
9 | };


--------------------------------------------------------------------------------
/src/interface/AbstractBinarizer.h:
--------------------------------------------------------------------------------
 1 | #include "../model/BinarizationResult.h"
 2 | #include "../model/RunConfiguration.h"
 3 | 
 4 | class AbstractBinarizer
 5 | {
 6 | public:
 7 |     virtual BinarizationResult *binarize(RunConfiguration *runConfig) = 0;
 8 | private:
 9 |     virtual MethodImplementation getBinarizerType() = 0;
10 | };


--------------------------------------------------------------------------------
/src/utils/CudaUtil.h:
--------------------------------------------------------------------------------
 1 | #include <string>
 2 | 
 3 | #pragma once
 4 | class CudaUtil
 5 | {
 6 | public:
 7 | 	static bool isGpuAvailable();
 8 | 	static bool setGpu(int deviceId);
 9 | 	static void getAvailableGpuNames();
10 | 	static std::string getDeviceName(int deviceId);
11 | 	static int getCurrentDevice();
12 | private: 
13 | 	CudaUtil();
14 | 	static int getDeviceIndexForName(std::string deviceName);
15 | };
16 | 
17 | 


--------------------------------------------------------------------------------
/src/model/ExecutionTimestamp.h:
--------------------------------------------------------------------------------
 1 | #include <string>
 2 | 
 3 | #pragma once
 4 | class ExecutionTimestamp
 5 | {
 6 | public:
 7 | 	ExecutionTimestamp();
 8 | 	~ExecutionTimestamp();
 9 | 	double histogramBuildingTimeInSeconds;
10 | 	double thresholdFindingTimeInSeconds;
11 | 	double binarizationTimeInSeconds;
12 | 	std::string toCommaSeparatedRow(std::string fileName, std::string tag);
13 | 	double getExecutionTime();
14 | };
15 | 
16 | 


--------------------------------------------------------------------------------
/src/model/BinarizationResult.h:
--------------------------------------------------------------------------------
 1 | #include <string>
 2 | #include "ExecutionTimestamp.h"
 3 | 
 4 | #pragma once
 5 | class BinarizationResult
 6 | {
 7 | private:
 8 |     unsigned int method;
 9 |     std::string binarizedImagePath;
10 |     ExecutionTimestamp *executionTimestamp;
11 | 
12 | public:
13 |     BinarizationResult(unsigned int method, std::string binarizedImagePath, ExecutionTimestamp* executionTimestamp);
14 |     ~BinarizationResult();
15 | 
16 |     void printResult();
17 | };
18 | 


--------------------------------------------------------------------------------
/src/core/binarizers/SMCudaOtsuBinarizer.cuh:
--------------------------------------------------------------------------------
 1 | #include "../../model/PngImage.h" 
 2 | #include "CudaOtsuBinarizer.cuh"
 3 | 
 4 | #pragma once
 5 | class SMCudaOtsuBinarizer : public CudaOtsuBinarizer
 6 | {
 7 | public:
 8 | 	SMCudaOtsuBinarizer(int threadsPerBlock, int numBlocks, bool drawHistogram);
 9 | 	~SMCudaOtsuBinarizer();
10 | protected:
11 | 	unsigned char cudaFindThreshold(double* histogram, long int totalPixels) override;
12 | 	double* cudaCalculateHistogram(unsigned char* rawPixels, long totalPixels) override;
13 | };
14 | 


--------------------------------------------------------------------------------
/src/core/binarizers/MonoCudaOtsuBinarizer.cuh:
--------------------------------------------------------------------------------
 1 | #include "../../model/PngImage.h" 
 2 | 
 3 | #pragma once
 4 | class MonoCudaOtsuBinarizer
 5 | {
 6 | public:
 7 | 	PngImage* binarize(PngImage* imageToBinarize);
 8 | 	MonoCudaOtsuBinarizer(int threadsPerBlock, bool drawHistogram, const char* TAG = "GPU - Single Kernel");
 9 | 	~MonoCudaOtsuBinarizer();
10 | protected:
11 | 	int threadsPerBlock_;
12 | 	float executionTime_;
13 | 	bool showHistogram_;
14 | 	const char* TAG;
15 | 	void showHistogram(double* histogram);
16 | 	unsigned char* cudaBinarize(unsigned char* rawPixels, long totalPixels);
17 | };
18 | 
19 | 


--------------------------------------------------------------------------------
/src/core/binarizers/OtsuBinarizer.h:
--------------------------------------------------------------------------------
 1 | #include "../../model/PngImage.h"
 2 | #include "../../interface/AbstractBinarizer.h" 
 3 | 
 4 | #pragma once
 5 | class OtsuBinarizer : public AbstractBinarizer
 6 | {
 7 | public:
 8 | 	BinarizationResult *binarize(RunConfiguration *runConfig);
 9 | 	OtsuBinarizer(PngImage* imageToBinarize);
10 | 	
11 | private:
12 | 	PngImage* imageToBinarize;
13 | 	std::vector<double> histogram;
14 | 
15 | 	MethodImplementation getBinarizerType();
16 | 	const char* getBinarizedFilePrefix();
17 | 	std::vector<double> calculateHistogram();
18 | 	int findThreshold();
19 | 	PngImage* binarize();
20 | 	void showHistogram();
21 | };
22 | 
23 | 


--------------------------------------------------------------------------------
/src/core/binarizers/OtsuOpenMPBinarizer.h:
--------------------------------------------------------------------------------
 1 | #include "../../model/PngImage.h"
 2 | #include "OtsuBinarizer.h"
 3 | 
 4 | #pragma once
 5 | class OtsuOpenMPBinarizer: public AbstractBinarizer
 6 | {
 7 | public:
 8 | 	BinarizationResult *binarize(RunConfiguration *runConfig);
 9 | 	OtsuOpenMPBinarizer(PngImage* imageToBinarize);
10 | 
11 | private:
12 | 	PngImage* imageToBinarize;
13 | 	std::vector<double> histogram;
14 | 	int cpuThreads;
15 | 
16 | 	MethodImplementation getBinarizerType();
17 | 	const char* getBinarizedFilePrefix();
18 | 	std::vector<double> calculateHistogram();
19 | 	int findThreshold();
20 | 	PngImage* binarize();
21 | 	void showHistogram();
22 | };


--------------------------------------------------------------------------------
/src/core/AppRunner.h:
--------------------------------------------------------------------------------
 1 | #include "../model/RunConfiguration.h"
 2 | #include "../model/BinarizationResult.h"
 3 | #include <map>
 4 | 
 5 | class AppRunner {
 6 | private:
 7 | 	static const int DEFAULT_THREADS_NUMBER = 512;
 8 | 	static const int DEFAULT_BLOCKS_NUMBER = 512;
 9 | 	static const int DEFAULT_CPU_THREADS = 16;
10 | 
11 |     RunConfiguration* runConfig;
12 |     std::map<MethodImplementation, BinarizationResult *> binarizationResults;
13 |     void printHelp();
14 |     int parseIntInputParam(const char *param, int defaultValue);
15 | public:
16 |     AppRunner();
17 |     ~AppRunner();
18 |     RunConfiguration* getRunConfig();
19 |     void loadInputConfiguration(int argc, char **argv);
20 |     std::map<MethodImplementation, BinarizationResult *> runBinarization();
21 | };


--------------------------------------------------------------------------------
/src/utils/RunConfigurationBuilder.h:
--------------------------------------------------------------------------------
 1 | #include <string>
 2 | #include "../model/PngImage.h"
 3 | #include "../model/RunConfiguration.h"
 4 | 
 5 | class RunConfigurationBuilder
 6 | {
 7 | private:
 8 | 	RunConfiguration* runConfiguration;
 9 | 
10 | public:
11 | 	RunConfigurationBuilder();
12 | 	RunConfigurationBuilder& forFileInPath(std::string fullFilePath);
13 | 	RunConfigurationBuilder& withThreadsPerBlock(int threadsPerBlock);
14 | 	RunConfigurationBuilder& withNumberOfBlocks(int numBlocks);
15 | 	RunConfigurationBuilder& withCpuThreads(int cpuThreads);
16 | 	RunConfigurationBuilder& withHistograms(bool drawHistograms);
17 | 	RunConfigurationBuilder& forImage(PngImage* loadedImage);
18 | 	RunConfigurationBuilder& withAlgorithmToRun(unsigned int alg);
19 | 	RunConfiguration* build();
20 | };


--------------------------------------------------------------------------------
/src/model/PngImage.h:
--------------------------------------------------------------------------------
 1 | #include <vector>
 2 | #pragma once
 3 | class PngImage
 4 | {
 5 | public:
 6 | 	PngImage(const char* filename, unsigned width, unsigned height, std::vector<unsigned char> rawPixelData);
 7 | 	~PngImage();
 8 | 	const char* getFilename();
 9 | 	unsigned getWidth();
10 | 	void setWidth(unsigned width);
11 | 	unsigned getHeight();
12 | 	void setHeight(unsigned height);
13 | 	long int getTotalPixels();
14 | 	std::vector<unsigned char> getRawPixelData();
15 | 	static const unsigned int MAX_PIXEL_VALUE = 256;
16 | 	static const unsigned char COLOR_BLACK = 0;
17 | 	static const unsigned char COLOR_WHITE = 255;
18 | private:
19 | 	const char* filename_;
20 | 	unsigned width_;
21 | 	unsigned height_;
22 | 	long int totalPixels_;
23 | 	std::vector<unsigned char> rawPixelData_;
24 | };
25 | 
26 | 


--------------------------------------------------------------------------------
/src/utils/ImageFileUtil.h:
--------------------------------------------------------------------------------
 1 | #include <vector>
 2 | #include <string>
 3 | #include "../model/PngImage.h"
 4 | 
 5 | #pragma once
 6 | class ImageFileUtil
 7 | {
 8 | public:
 9 | 	static PngImage* loadPngFile(const char* filename);
10 | 	static void savePngFile(PngImage* pngImage, const char* newFileName);
11 | 	static std::string addPrefix(std::string fullFilePath, const char* prefix);
12 | 	static std::string joinString(std::vector<std::string> strings, const char delimiter = '\0');
13 | 	static void saveCsvFile(std::vector<std::string> rows, const char* filename);
14 | private:
15 | 	ImageFileUtil();
16 | 	static std::vector<std::string> splitString(std::string stringToSplit, const char delimiter);
17 | 	static bool fileExists(const char * fileName);
18 | 	static char getOsPathDelimiter();
19 | };
20 | 
21 | 


--------------------------------------------------------------------------------
/src/model/RunConfiguration.h:
--------------------------------------------------------------------------------
 1 | #pragma once
 2 | #include <string>
 3 | #include "PngImage.h"
 4 | #include "../interface/MethodImplementation.h"
 5 | 
 6 | class RunConfiguration
 7 | {
 8 | private:
 9 | 	std::string fullFilePath;
10 | 	int threadsPerBlock;
11 | 	int numBlocks;
12 | 	int cpuThreads;
13 | 	bool drawHistograms;
14 | 	PngImage* loadedImage;
15 | 
16 |     bool algChosenToRun[6] = {false, false, false, false, false, false};
17 | 
18 | public:
19 | 	RunConfiguration();
20 | 	~RunConfiguration();
21 | 	friend class RunConfigurationBuilder;
22 | 
23 | 	std::string getFullFilePath();
24 | 	int getThreadsPerBlock();
25 | 	int getNumberOfBlocks();
26 | 	int getCpuThreads();
27 | 	bool shouldDrawHistograms();
28 | 	PngImage* getLoadedImage();
29 | 	bool hasLoadedImage();
30 | 	bool shouldRunAlgorithm(unsigned int algorithm);
31 | 	void print();
32 | };


--------------------------------------------------------------------------------
/src/core/binarizers/CudaOtsuBinarizer.cuh:
--------------------------------------------------------------------------------
 1 | #include "../../model/PngImage.h"
 2 | #include "../../model/ExecutionTimestamp.h"
 3 | 
 4 | #pragma once
 5 | class CudaOtsuBinarizer
 6 | {
 7 | public:
 8 | 	PngImage* binarize(PngImage* imageToBinarize);
 9 | 	std::string getBinarizerExecutionInfo(std::string fileName);
10 | 	CudaOtsuBinarizer(int threadsPerBlock, int numBlocks, bool drawHistogram, const char* TAG = "GPU");
11 | 	virtual ~CudaOtsuBinarizer();
12 | protected:
13 | 	int threadsPerBlock_;
14 | 	int numBlocks_;
15 | 	ExecutionTimestamp* binarizerTimestamp_;
16 | 	bool drawHistogram_;
17 | 	const char* TAG;
18 | 	virtual void showHistogram(double* histogram);
19 | 	virtual double* cudaCalculateHistogram(unsigned char* rawPixels, long totalPixels);
20 | 	virtual unsigned char cudaFindThreshold(double* histogram, long int totalPixels);
21 | 	virtual unsigned char* cudaBinarize(unsigned char* rawPixels, long totalPixels, unsigned char threshold);
22 | };
23 | 
24 | 


--------------------------------------------------------------------------------
/Makefile:
--------------------------------------------------------------------------------
 1 | SOURCE_DIR = src
 2 | BUILD_DIR = build
 3 | EXEC_FILE = CudaOtsu
 4 | DEFAULT_VALUE_FLAG = -1
 5 | 
 6 | SOURCE_FILES := 
 7 | ifeq ($(OS), Windows_NT)
 8 | 	SOURCE_FILES = $(shell find $(SOURCE_DIR) -name *.cpp -o -name *.cu)
 9 | else 
10 | 	ifeq ($(UNAME_S), Linux)
11 | 		SOURCE_FILES = $(shell find $(SOURCE_DIR) -name '*.cpp' -o -name '*.cu')
12 | 	else
13 | 		SOURCE_FILES = None
14 | 	endif
15 | endif
16 | 
17 | list_sources:
18 | 	@echo "Source files:"
19 | 	@echo ${SOURCE_FILES}
20 | 
21 | build:
22 | 	mkdir -p ${BUILD_DIR}
23 | 	nvcc -x cu ${SOURCE_FILES} --std=c++11 -lineinfo -o ${BUILD_DIR}/${EXEC_FILE} -Xcompiler -openmp
24 | 
25 | run:
26 | 	./${BUILD_DIR}/${EXEC_FILE} $(file) $(threads) $(blocks) -d $(device_id)
27 | 
28 | run_default:
29 | 	./${BUILD_DIR}/${EXEC_FILE} $(file) ${DEFAULT_VALUE_FLAG} ${DEFAULT_VALUE_FLAG} -d 0
30 | 
31 | run_histogram:
32 | 	./${BUILD_DIR}/${EXEC_FILE} $(file) $(threads) $(blocks) -d $(device_id) -h
33 | 
34 | clean:
35 | 	rm -rf ${BUILD_DIR}
36 | 


--------------------------------------------------------------------------------
/src/model/PngImage.cpp:
--------------------------------------------------------------------------------
 1 | #include "PngImage.h"
 2 | 
 3 | 
 4 | 
 5 | PngImage::PngImage(const char* filename, unsigned width, unsigned height, std::vector<unsigned char> rawPixelData)
 6 | {
 7 | 	this->filename_ = filename;
 8 | 	this->width_ = width;
 9 | 	this->height_ = height;
10 | 	this->rawPixelData_ = rawPixelData;
11 | 	this->totalPixels_ = rawPixelData.size();
12 | }
13 | 
14 | PngImage::~PngImage()
15 | {
16 | 	this->rawPixelData_.clear();
17 | 	this->rawPixelData_.resize(0);
18 | 	this->rawPixelData_.shrink_to_fit();
19 | }
20 | 
21 | const char * PngImage::getFilename()
22 | {
23 | 	return filename_;
24 | }
25 | 
26 | unsigned PngImage::getWidth()
27 | {
28 | 	return width_;
29 | }
30 | 
31 | void PngImage::setWidth(unsigned width)
32 | {
33 | 	this->width_ = width;
34 | }
35 | 
36 | unsigned PngImage::getHeight()
37 | {
38 | 	return height_;
39 | }
40 | 
41 | void PngImage::setHeight(unsigned height)
42 | {
43 | 	this->height_ = height;
44 | }
45 | 
46 | long int PngImage::getTotalPixels()
47 | {
48 | 	return totalPixels_;
49 | }
50 | 
51 | std::vector<unsigned char> PngImage::getRawPixelData()
52 | {
53 | 	return rawPixelData_;
54 | }
55 | 


--------------------------------------------------------------------------------
/LICENSE:
--------------------------------------------------------------------------------
 1 | MIT License
 2 | 
 3 | Copyright (c) 2020 Dawid Paluchowski
 4 | 
 5 | Permission is hereby granted, free of charge, to any person obtaining a copy
 6 | of this software and associated documentation files (the "Software"), to deal
 7 | in the Software without restriction, including without limitation the rights
 8 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
 9 | copies of the Software, and to permit persons to whom the Software is
10 | furnished to do so, subject to the following conditions:
11 | 
12 | The above copyright notice and this permission notice shall be included in all
13 | copies or substantial portions of the Software.
14 | 
15 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
18 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21 | SOFTWARE.
22 | 


--------------------------------------------------------------------------------
/src/model/ExecutionTimestamp.cpp:
--------------------------------------------------------------------------------
 1 | #include "ExecutionTimestamp.h"
 2 | 
 3 | #include "../utils/ImageFileUtil.h"
 4 | #include <vector>
 5 | 
 6 | ExecutionTimestamp::ExecutionTimestamp()
 7 | {
 8 | 	this->histogramBuildingTimeInSeconds = 0;
 9 | 	this->thresholdFindingTimeInSeconds = 0;
10 | 	this->binarizationTimeInSeconds = 0;
11 | }
12 | 
13 | 
14 | ExecutionTimestamp::~ExecutionTimestamp()
15 | {
16 | }
17 | 
18 | // Format: fileName,TAG,histogramBuildingTime,thresholdFindingTime,binarizationTime,executionTime
19 | std::string ExecutionTimestamp::toCommaSeparatedRow(std::string fileName, std::string tag)
20 | {
21 | 	std::vector<std::string> values;
22 | 	values.push_back(fileName);
23 | 	values.push_back(tag);
24 | 	values.push_back(std::to_string(histogramBuildingTimeInSeconds));
25 | 	values.push_back(std::to_string(thresholdFindingTimeInSeconds));
26 | 	values.push_back(std::to_string(binarizationTimeInSeconds));
27 | 	values.push_back(std::to_string(getExecutionTime()));
28 | 
29 | 	return ImageFileUtil::joinString(values, ',');
30 | }
31 | 
32 | double ExecutionTimestamp::getExecutionTime()
33 | {
34 | 	return histogramBuildingTimeInSeconds + thresholdFindingTimeInSeconds + binarizationTimeInSeconds;
35 | }
36 | 


--------------------------------------------------------------------------------
/src/model/BinarizationResult.cpp:
--------------------------------------------------------------------------------
 1 | #include "BinarizationResult.h"
 2 | 
 3 | BinarizationResult::BinarizationResult(unsigned int method, std::string binarizedImagePath, ExecutionTimestamp* executionTimestamp)
 4 | {
 5 |     this->method = method;
 6 |     this->binarizedImagePath = binarizedImagePath;
 7 |     this->executionTimestamp = executionTimestamp;
 8 | }
 9 | 
10 | 
11 | BinarizationResult::~BinarizationResult() {};
12 | 
13 | void BinarizationResult::printResult()
14 | {
15 |     printf("\n---------------------------------------------\n");
16 |     printf("\nBinarization result:\n");
17 | 	printf("\tMethod: %d\n", this->method);
18 | 	printf("\tBinarized image path: %s\n", this->binarizedImagePath.c_str());
19 | 
20 |     if (this->executionTimestamp->histogramBuildingTimeInSeconds != 0) {
21 | 	    printf("\tExecution - histogram build time: %f seconds\n", this->executionTimestamp->histogramBuildingTimeInSeconds);
22 |     }
23 | 
24 |     if (this->executionTimestamp->thresholdFindingTimeInSeconds != 0) {
25 | 	    printf("\tExecution - threshold lookup time: %f seconds\n", this->executionTimestamp->thresholdFindingTimeInSeconds);
26 |     }
27 | 
28 |     if (this->executionTimestamp->binarizationTimeInSeconds != 0) {
29 | 	    printf("\tExecution - binarization time: %f seconds\n", this->executionTimestamp->binarizationTimeInSeconds);
30 |     }
31 |     printf("\n---------------------------------------------\n");
32 | }


--------------------------------------------------------------------------------
/src/model/RunConfiguration.cpp:
--------------------------------------------------------------------------------
 1 | #include "RunConfiguration.h"
 2 | 
 3 | RunConfiguration::RunConfiguration() {}
 4 | 
 5 | RunConfiguration::~RunConfiguration()
 6 | {
 7 | 	delete this->loadedImage;
 8 | }
 9 | 
10 | std::string RunConfiguration::getFullFilePath()
11 | {
12 | 	return this->fullFilePath;
13 | }
14 | 
15 | int RunConfiguration::getThreadsPerBlock()
16 | {
17 | 	return this->threadsPerBlock;
18 | }
19 | 
20 | int RunConfiguration::getNumberOfBlocks()
21 | {
22 | 	return this->numBlocks;
23 | }
24 | 
25 | int RunConfiguration::getCpuThreads()
26 | {
27 | 	return this->cpuThreads;
28 | }
29 | 
30 | bool RunConfiguration::shouldDrawHistograms()
31 | {
32 | 	return this->drawHistograms;
33 | }
34 | 
35 | PngImage *RunConfiguration::getLoadedImage()
36 | {
37 | 	return this->loadedImage;
38 | }
39 | 
40 | bool RunConfiguration::hasLoadedImage()
41 | {
42 | 	return this->loadedImage != nullptr;
43 | }
44 | 
45 | bool RunConfiguration::shouldRunAlgorithm(unsigned int algorithm)
46 | {
47 | 	return this->algChosenToRun[algorithm] || this->algChosenToRun[ALL];
48 | }
49 | 
50 | void RunConfiguration::print()
51 | {
52 | 	printf("\nRun configuration:\n");
53 | 	printf("Full file path: %s\n", this->getFullFilePath().c_str());
54 | 	printf("Number of blocks: %d\n", this->getNumberOfBlocks());
55 | 	printf("Threads per block: %d\n", this->getThreadsPerBlock());
56 | 	printf("CPU threads: %d\n", this->getCpuThreads());
57 | 	printf("Should draw histograms: %d\n", this->shouldDrawHistograms());
58 | 	printf("Image loaded: %d\n", this->hasLoadedImage());
59 | }


--------------------------------------------------------------------------------
/src/utils/RunConfigurationBuilder.cpp:
--------------------------------------------------------------------------------
 1 | #include "RunConfigurationBuilder.h"
 2 | 
 3 | RunConfigurationBuilder::RunConfigurationBuilder() 
 4 | {
 5 | 	this->runConfiguration = new RunConfiguration();
 6 | }
 7 | 
 8 | RunConfigurationBuilder &RunConfigurationBuilder::forFileInPath(std::string fullFilePath)
 9 | {
10 | 	runConfiguration->fullFilePath = fullFilePath;
11 | 	return *this;
12 | }
13 | 
14 | RunConfigurationBuilder &RunConfigurationBuilder::withThreadsPerBlock(int threadsPerBlock)
15 | {
16 | 	runConfiguration->threadsPerBlock = threadsPerBlock;
17 | 	return *this;
18 | }
19 | 
20 | RunConfigurationBuilder &RunConfigurationBuilder::withNumberOfBlocks(int numBlocks)
21 | {
22 | 	runConfiguration->numBlocks = numBlocks;
23 | 	return *this;
24 | }
25 | 
26 | RunConfigurationBuilder &RunConfigurationBuilder::withCpuThreads(int cpuThreads)
27 | {
28 | 	runConfiguration->cpuThreads = cpuThreads;
29 | 	return *this;
30 | }
31 | 
32 | RunConfigurationBuilder &RunConfigurationBuilder::withHistograms(bool drawHistograms)
33 | {
34 | 	runConfiguration->drawHistograms = drawHistograms;
35 | 	return *this;
36 | }
37 | 
38 | RunConfigurationBuilder &RunConfigurationBuilder::forImage(PngImage *loadedImage)
39 | {
40 | 	runConfiguration->loadedImage = loadedImage;
41 | 	return *this;
42 | }
43 | 
44 | RunConfigurationBuilder &RunConfigurationBuilder::withAlgorithmToRun(unsigned int alg)
45 | {
46 | 	runConfiguration->algChosenToRun[alg] = true;
47 | 	return *this;
48 | }
49 | 
50 | RunConfiguration *RunConfigurationBuilder::build()
51 | {
52 | 	return this->runConfiguration;
53 | }


--------------------------------------------------------------------------------
/src/utils/CudaUtil.cpp:
--------------------------------------------------------------------------------
 1 | #include "CudaUtil.h"
 2 | #include <string.h>
 3 | 
 4 | // CUDA imports
 5 | #include <cuda_runtime.h>
 6 | 
 7 | bool CudaUtil::isGpuAvailable()
 8 | {
 9 | 	int devicesCount;
10 | 	cudaGetDeviceCount(&devicesCount);
11 | 	for (int deviceIndex = 0; deviceIndex < devicesCount; deviceIndex++) {
12 | 		cudaDeviceProp deviceProperties;
13 | 		cudaGetDeviceProperties(&deviceProperties, deviceIndex);
14 | 
15 | 		if (deviceProperties.major >= 2
16 | 			&& deviceProperties.minor >= 0)
17 | 		{
18 | 			return true;
19 | 		}
20 | 	}
21 | 
22 | 	return false;
23 | }
24 | 
25 | void CudaUtil::getAvailableGpuNames()
26 | {
27 | 	int devicesCount;
28 | 	cudaGetDeviceCount(&devicesCount);
29 | 	printf("\nAvailable GPUs:\n");
30 | 	for (int deviceIndex = 0; deviceIndex < devicesCount; ++deviceIndex)
31 | 	{
32 | 		cudaDeviceProp deviceProperties;
33 | 		cudaGetDeviceProperties(&deviceProperties, deviceIndex);
34 | 		printf("[%d] %s\n", deviceIndex, deviceProperties.name);
35 | 	}
36 | }
37 | 
38 | std::string CudaUtil::getDeviceName(int deviceId)
39 | {
40 | 	char* deviceName;
41 | 	int devicesCount;
42 | 	cudaGetDeviceCount(&devicesCount);
43 | 	if (deviceId < devicesCount) {
44 | 		cudaDeviceProp deviceProperties;
45 | 		cudaGetDeviceProperties(&deviceProperties, deviceId);
46 | 		deviceName = deviceProperties.name;
47 | 	} else {
48 | 		deviceName = "-";
49 | 	}
50 | 
51 | 	return std::string(deviceName);
52 | }
53 | 
54 | int CudaUtil::getCurrentDevice()
55 | {
56 | 	int deviceId;
57 | 	cudaGetDevice(&deviceId);
58 | 
59 | 	return deviceId;
60 | }
61 | 
62 | bool CudaUtil::setGpu(int deviceIndex)
63 | {
64 | 	int devicesCount;
65 | 	cudaGetDeviceCount(&devicesCount);
66 | 	if (deviceIndex < devicesCount) {
67 | 		cudaDeviceProp deviceProperties;
68 | 		cudaGetDeviceProperties(&deviceProperties, deviceIndex);
69 | 		printf("Selected GPU: (%d) - %s\n", deviceIndex, deviceProperties.name);
70 | 		printf("Compute Capability: %d.%d\n", deviceProperties.major, deviceProperties.minor);
71 | 		cudaSetDevice(deviceIndex);
72 | 		return true;
73 | 	} else {
74 | 		printf("No GPU available for given index: %d\n", deviceIndex);
75 | 		return false;
76 | 	}
77 | }
78 | 
79 | CudaUtil::CudaUtil() {}
80 | 
81 | int CudaUtil::getDeviceIndexForName(std::string deviceName)
82 | {
83 | 	int devicesCount;
84 | 	cudaGetDeviceCount(&devicesCount);
85 | 	for (int deviceIndex = 0; deviceIndex < devicesCount; ++deviceIndex)
86 | 	{
87 | 		cudaDeviceProp deviceProperties;
88 | 		cudaGetDeviceProperties(&deviceProperties, deviceIndex);
89 | 		if (deviceProperties.name == deviceName)
90 | 		{
91 | 			return deviceIndex;
92 | 		}
93 | 	}
94 | 
95 | 	return -1;
96 | }
97 | 


--------------------------------------------------------------------------------
/src/utils/ImageFileUtil.cpp:
--------------------------------------------------------------------------------
  1 | #include "ImageFileUtil.h"
  2 | #include "../libs/lodepng.h"
  3 | #include <iostream>
  4 | #include <sstream>
  5 | #include <fstream>
  6 | 
  7 | ImageFileUtil::ImageFileUtil() {}
  8 | 
  9 | PngImage* ImageFileUtil::loadPngFile(const char* filename) {
 10 | 	std::vector<unsigned char> png;
 11 | 	std::vector<unsigned char> rawImage;
 12 | 
 13 | 	unsigned imageWidth;
 14 | 	unsigned imageHeight;
 15 | 
 16 | 	if (!fileExists(filename)) {
 17 | 		std::cout << "Cannot find or open file: " << filename << std::endl;
 18 | 		return nullptr;
 19 | 	}
 20 | 
 21 | 	unsigned error = lodepng::load_file(png, filename);
 22 | 	if (!error) {
 23 | 		error = lodepng::decode(rawImage, imageWidth, imageHeight, png);
 24 | 	}
 25 | 
 26 | 	if (error) {
 27 | 		std::cout << lodepng_error_text(error) << std::endl;
 28 | 		return nullptr;
 29 | 	}
 30 | 
 31 | 	return new PngImage(filename, imageWidth, imageHeight, rawImage);
 32 | }
 33 | 
 34 | void ImageFileUtil::savePngFile(PngImage* pngImage, const char* newFileName = nullptr) {
 35 | 	std::vector<unsigned char> png;
 36 | 
 37 | 	unsigned error = lodepng::encode(png, pngImage->getRawPixelData(), pngImage->getWidth(), pngImage->getHeight());
 38 | 
 39 | 	const char* filename = newFileName != nullptr ? newFileName : pngImage->getFilename();
 40 | 
 41 | 	if (!error) lodepng::save_file(png, filename);
 42 | 
 43 | 	if (error) {
 44 | 		std::cout << lodepng_error_text(error) << std::endl;
 45 | 	}
 46 | }
 47 | 
 48 | std::string ImageFileUtil::addPrefix(std::string fullFilePath, const char* prefix) {
 49 | 	std::vector<std::string> pathParts;
 50 | 
 51 | 	const char osPathDelimiter = getOsPathDelimiter();
 52 | 	
 53 | 	pathParts = splitString(fullFilePath, osPathDelimiter);
 54 | 
 55 | 	std::string newPathPart = pathParts.back();
 56 | 	pathParts.pop_back();
 57 | 	pathParts.push_back(prefix + newPathPart);
 58 | 
 59 | 	return joinString(pathParts, osPathDelimiter);
 60 | }
 61 | 
 62 | std::vector<std::string> ImageFileUtil::splitString(std::string stringToSplit, const char delimiter) {
 63 | 	std::vector<std::string> parts;
 64 | 	std::istringstream f(stringToSplit);
 65 | 	std::string part;
 66 | 	while (std::getline(f, part, delimiter)) {
 67 | 		parts.push_back(part);
 68 | 	}
 69 | 	
 70 | 	return parts;
 71 | }
 72 | 
 73 | std::string ImageFileUtil::joinString(std::vector<std::string> strings, const char delimiter) {
 74 | 	std::string resultString = strings.front();
 75 | 	for (std::vector<std::string>::size_type i = 1; i != strings.size(); i++) {
 76 | 		resultString.append(delimiter + strings[i]);
 77 | 	}
 78 | 
 79 | 	return resultString;
 80 | }
 81 | 
 82 | void ImageFileUtil::saveCsvFile(std::vector<std::string> rows, const char * filename)
 83 | {
 84 | 	std::ofstream csvFile;
 85 | 	csvFile.open(filename, std::ofstream::app);
 86 | 
 87 | 	for (std::vector<std::string>::size_type i = 0; i != rows.size(); i++) {
 88 | 		csvFile << rows[i] + "\n";
 89 | 	}
 90 | 
 91 | 	csvFile.close();
 92 | }
 93 | 
 94 | bool ImageFileUtil::fileExists(const char *fileName)
 95 | {
 96 | 	std::ifstream infile(fileName);
 97 | 	return infile.good();
 98 | }
 99 | 
100 | char ImageFileUtil::getOsPathDelimiter() {
101 | #if defined(WIN32) || defined(_WIN32) || defined(__WIN32) && !defined(__CYGWIN__)
102 | 	return '\\';
103 | #endif
104 | 	return '/';
105 | }
106 | 
107 | 


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
 1 | # CudaOtsu
 2 | Otsu's method thresholding and image binarization on GPU using CUDA in C++.
 3 | 
 4 | ## Otsu's Method
 5 | Simple and one of the most popular image thresholding method used in computer vision problems. Algorithm helps to find optimal threshold value for greyscale image to be then used in image binarization. As binarized images are main data type for problems such as OCR, Otsu's method is used as a part of pre-processing pipelines for computer vision problems.
 6 | 
 7 | Otsu's main idea is to find threshold that minimzes the intra-class variance within 'foreground' and 'background' classes. Based on observation that minimazing intra-class variance is the same as maximizing inter-class variance, we can define algorithm as following steps:
 8 | 
 9 | 1. Compute histogram and probabilities of each intensity level
10 | 2. Set up initial classes probability and classes mean
11 | 3. For every intensity level:
12 |     - Update class probability and mean for current intensity level
13 |     - Compute inter-class variance
14 | 4. Threshold will be at intenisty level with highest inter-class variance
15 | 
16 | ### Otsu's Method Visualization
17 | ![Otsu's Method Visualization](https://upload.wikimedia.org/wikipedia/commons/3/34/Otsu%27s_Method_Visualization.gif)
18 | 
19 | ## Project goal
20 | Project is focused on implementing Otsu's method as CUDA kernels to test how well GPU will handle this algorithm in terms of computation time (comparing to multithreaded CPU implementation). Also it's a kind of CUDA playground for me as I'm absolute beginner in GPU computations. 
21 | 
22 | ## Current features 
23 | - C++ implementation of Otsu's method on CPU (single threaded)
24 | - Basic CUDA implementation of Otsu's method on GPU
25 | - Basic CUDA shared memory usage (no huge speed boost here, Otsu's algorithm gains very little from cache)
26 | - Makefile for more multiplatform approach
27 | - Extendable Binarizers architecture
28 | 
29 | ## How to run project
30 | - Build project using makefile build target or visual studio project build  
31 | ```bash
32 | $> make clean
33 | $> make build
34 | ```
35 | If you have any errors during `make build` command, please check if the SOURCE_FILES variable (in Makefile) is set correctly according to your OS!
36 | 
37 | - Run it using makefile
38 | ```bash
39 | $> make run 
40 |     file=<absolute_path_to_file>
41 |     threads=<cuda_threads_number>
42 |     blocks=<cuda_blocks_number>
43 |     device_id=<gpu_id> 
44 |     # default GPU indexed as 0, for more info use nvidia-smi tool
45 | ```
46 | 
47 | - or directly from executable (`.exe` in case of Windows OS)
48 | ```bash
49 | $> ./cudaOtsu <absolute_path_to_file> <cuda_threads_number>  
50 |     <cuda_blocks_number> -d <gpu_id> [optional flags]
51 |     # Flags:
52 |     #   -h (show histogram values for each binarizer run)
53 |     #   --cpu (run CPU implementation)
54 |     #   --gpu (run basic GPU implementation)
55 |     #   --gpu-sm (run shared memory optimized GPU implementation)
56 |     #   --gpu-mono (run GPU version with singlekernel architecture on single GPU block)
57 |     #   --run-all (run all implemented versions of Otsu algorithm both CPU and GPU)
58 | ```
59 | 
60 | ## To do
61 | - [ ] Concurrent CPU method implementation for benchmarking (openMP)
62 | - [ ] Memory optimization
63 | - [ ] Simple (cross-platform) GUI 
64 | - [ ] Cuda GPU selection error handling
65 | - [ ] Research and compare different CUDA optimization mechanisms 
66 | 


--------------------------------------------------------------------------------
/src/core/binarizers/OtsuBinarizer.cpp:
--------------------------------------------------------------------------------
  1 | #include "../../utils/ImageFileUtil.h"
  2 | #include "OtsuBinarizer.h"
  3 | #include <stdio.h>
  4 | #include <time.h>
  5 | 
  6 | OtsuBinarizer::OtsuBinarizer(PngImage* imageToBinarize)
  7 | {
  8 | 	this->imageToBinarize = imageToBinarize;
  9 | 	this->histogram = calculateHistogram();
 10 | }
 11 | 
 12 | MethodImplementation OtsuBinarizer::getBinarizerType()
 13 | {
 14 | 	return CPU;
 15 | }
 16 | 
 17 | const char* OtsuBinarizer::getBinarizedFilePrefix()
 18 | {
 19 | 	return "cpu_binarized_";
 20 | }
 21 | 
 22 | BinarizationResult* OtsuBinarizer::binarize(RunConfiguration* runConfig)
 23 | {
 24 | 	ExecutionTimestamp* executionTimestamp = new ExecutionTimestamp();
 25 | 	clock_t time;
 26 | 	time = clock();
 27 | 
 28 | 	std::string cpuBinarizedFilename = ImageFileUtil::addPrefix(runConfig->getFullFilePath(), getBinarizedFilePrefix());
 29 | 
 30 | 	PngImage *cpuBinarizedImage = binarize();
 31 | 
 32 | 	time = clock() - time;
 33 | 
 34 | 	executionTimestamp->binarizationTimeInSeconds = ((double)time / CLOCKS_PER_SEC);
 35 | 
 36 | 	ImageFileUtil::savePngFile(cpuBinarizedImage, cpuBinarizedFilename.c_str());
 37 | 
 38 | 	delete cpuBinarizedImage;
 39 | 
 40 | 	return new BinarizationResult(
 41 | 		getBinarizerType(),
 42 | 		cpuBinarizedFilename.c_str(),
 43 | 		executionTimestamp
 44 | 	);
 45 | }
 46 | 
 47 | std::vector<double> OtsuBinarizer::calculateHistogram() 
 48 | {
 49 | 	std::vector<double> histogram(PngImage::MAX_PIXEL_VALUE);
 50 | 	std::vector<unsigned char> image = this->imageToBinarize->getRawPixelData();
 51 | 
 52 | 	std::vector<unsigned char> occurences(PngImage::MAX_PIXEL_VALUE);
 53 | 	unsigned char pixelValue;
 54 | 	long totalPixels = image.size();
 55 | 
 56 | 	for (std::vector<unsigned char>::size_type i = 0; i != totalPixels; i++) {
 57 | 		pixelValue = image[i];
 58 | 		histogram[pixelValue]++;
 59 | 	}
 60 | 
 61 | 	// Normalization
 62 | 	for (std::vector<unsigned char>::size_type v = 0; v != PngImage::MAX_PIXEL_VALUE; v++) {
 63 | 		histogram[v] /= totalPixels;
 64 | 	}
 65 | 
 66 | 	return histogram;
 67 | }
 68 | 
 69 | PngImage* OtsuBinarizer::binarize() {
 70 | 	int threshold = findThreshold();
 71 | 	std::vector<unsigned char> imagePixels = this->imageToBinarize->getRawPixelData();
 72 | 	for (std::vector<unsigned char>::size_type i = 0; i != imageToBinarize->getTotalPixels(); i++) {
 73 | 		if ((int)imagePixels[i] > threshold) {
 74 | 			imagePixels[i] = PngImage::COLOR_WHITE;
 75 | 		} else {
 76 | 			imagePixels[i] = PngImage::COLOR_BLACK;
 77 | 		}
 78 | 	}
 79 | 
 80 | 	return new PngImage(
 81 | 		imageToBinarize->getFilename(),
 82 | 		imageToBinarize->getWidth(),
 83 | 		imageToBinarize->getHeight(),
 84 | 		imagePixels
 85 | 	);
 86 | }
 87 | 
 88 | int OtsuBinarizer::findThreshold() {
 89 | 	int threshold;
 90 | 	long int totalPixels = this->imageToBinarize->getTotalPixels();
 91 | 	double firstClassProbability = 0, secondClassProbability = 0;
 92 | 	double firstClassMean = 0, secondClassMean = 0;
 93 | 	double betweenClassVariance = 0, maxVariance = 0;
 94 | 	double allProbabilitySum = 0, firstProbabilitySum = 0;
 95 | 
 96 | 	for (int i = 0; i < PngImage::MAX_PIXEL_VALUE; i++) {
 97 | 		allProbabilitySum += i * histogram[i];
 98 | 	}
 99 | 
100 | 	for (int t = 0; t < PngImage::MAX_PIXEL_VALUE; t++) {
101 | 		firstClassProbability += histogram[t];
102 | 		secondClassProbability = 1 - firstClassProbability;
103 | 
104 | 		firstProbabilitySum += t * histogram[t];
105 | 		firstClassMean = (double)firstProbabilitySum / (double)firstClassProbability;
106 | 		secondClassMean = (double)(allProbabilitySum - firstProbabilitySum) / (double)secondClassProbability;
107 | 
108 | 		betweenClassVariance = firstClassProbability * secondClassProbability * pow((firstClassMean - secondClassMean), 2);
109 | 
110 | 		if (betweenClassVariance > maxVariance) {
111 | 			threshold = t;
112 | 			maxVariance = betweenClassVariance;
113 | 		}
114 | 	}
115 | 
116 | 	printf("[CPU] Threshold value: %d", threshold);
117 | 
118 | 	return threshold;
119 | }
120 | 
121 | void OtsuBinarizer::showHistogram() {
122 | 	printf("\nHistogram:\n");
123 | 	double value = 0;
124 | 	for (int i = 0; i < PngImage::MAX_PIXEL_VALUE; i++) {
125 | 		value = histogram[i];
126 | 		printf("\tPixel value %d -> %.5f\n", i, value);
127 | 	}
128 | }
129 | 


--------------------------------------------------------------------------------
/src/core/AppRunner.cpp:
--------------------------------------------------------------------------------
  1 | #include "AppRunner.h"
  2 | #include "../utils/CudaUtil.h"
  3 | #include "../utils/ImageFileUtil.h"
  4 | #include "../utils/RunConfigurationBuilder.h"
  5 | 
  6 | AppRunner::AppRunner() {}
  7 | 
  8 | void AppRunner::loadInputConfiguration(int argc, char **argv)
  9 | {
 10 | 
 11 |     RunConfigurationBuilder configBuilder = RunConfigurationBuilder();
 12 | 
 13 |     std::string fullFilePath;
 14 |     int cudaDeviceId;
 15 | 
 16 |     if (argc <= 3)
 17 |     {
 18 |         printHelp();
 19 |         CudaUtil::getAvailableGpuNames();
 20 |     }
 21 |     else
 22 |     {
 23 |         fullFilePath = argv[1];
 24 |         configBuilder.forFileInPath(fullFilePath);
 25 |         configBuilder.withThreadsPerBlock(parseIntInputParam(argv[2], DEFAULT_THREADS_NUMBER));
 26 |         configBuilder.withNumberOfBlocks(parseIntInputParam(argv[3], DEFAULT_BLOCKS_NUMBER));
 27 |         configBuilder.withCpuThreads(parseIntInputParam(argv[4], DEFAULT_CPU_THREADS));
 28 |         configBuilder.withHistograms(false);
 29 | 
 30 |         for (int argumentIndex = 5; argumentIndex < argc; argumentIndex++)
 31 |         {
 32 |             std::string flag(argv[argumentIndex]);
 33 | 
 34 |             if (flag == "-h")
 35 |             {
 36 |                 configBuilder.withHistograms(true);
 37 |                 continue;
 38 |             }
 39 | 
 40 |             if (flag == "-d")
 41 |             {
 42 |                 int nextArgument = argumentIndex + 1;
 43 |                 if (nextArgument < argc)
 44 |                 {
 45 |                     cudaDeviceId = std::atoi(argv[nextArgument]);
 46 | 
 47 |                     bool gpuSetSuccess = CudaUtil::setGpu(cudaDeviceId);
 48 | 
 49 |                     if (!gpuSetSuccess)
 50 |                     {
 51 |                         CudaUtil::getAvailableGpuNames();
 52 |                     }
 53 | 
 54 |                     argumentIndex = nextArgument;
 55 |                     continue;
 56 |                 }
 57 |             }
 58 | 
 59 |             if (flag == "--cpu")
 60 |             {
 61 |                 configBuilder.withAlgorithmToRun(CPU);
 62 |                 continue;
 63 |             }
 64 | 
 65 |             if (flag == "--cpu-openmp")
 66 |             {
 67 |                 configBuilder.withAlgorithmToRun(CPU_OpenMP);
 68 |                 continue;
 69 |             }
 70 | 
 71 |             if (flag == "--gpu")
 72 |             {
 73 |                 configBuilder.withAlgorithmToRun(GPU);
 74 |                 continue;
 75 |             }
 76 | 
 77 |             if (flag == "--gpu-sm")
 78 |             {
 79 |                 configBuilder.withAlgorithmToRun(GPU_SharedMemory);
 80 |                 continue;
 81 |             }
 82 | 
 83 |             if (flag == "--gpu-mono")
 84 |             {
 85 |                 configBuilder.withAlgorithmToRun(GPU_MonoKernel);
 86 |                 continue;
 87 |             }
 88 | 
 89 |             if (flag == "--run-all")
 90 |             {
 91 |                 configBuilder.withAlgorithmToRun(ALL);
 92 |                 continue;
 93 |             }
 94 |         }
 95 |     }
 96 | 
 97 |      this->runConfig = configBuilder
 98 |                           .forImage(ImageFileUtil::loadPngFile(fullFilePath.c_str()))
 99 |                           .build();
100 | }
101 | 
102 | RunConfiguration* AppRunner::getRunConfig()
103 | {
104 |     return this->runConfig;
105 | }
106 | 
107 | void AppRunner::printHelp()
108 | {
109 |     std::string helpMessage = "";
110 |     helpMessage.append("Help:\n");
111 |     helpMessage.append("<program> filePath cudaThreadsNumber cudaBlocksNumber [optional flags]\n");
112 |     helpMessage.append("\tFlags:\n");
113 |     helpMessage.append("\t\t -h show histogram values for each binarizer run\n");
114 |     helpMessage.append("\t\t -d <deviceName> choose GPU device by given name (defaults to 0)\n");
115 |     helpMessage.append("\t\t --cpu run CPU version of algorithm\n");
116 |     helpMessage.append("\t\t --cpu-openmp run CPU with OpenMP version of algorithm\n");
117 |     helpMessage.append("\t\t --gpu run GPU reference version of algorithm\n");
118 |     helpMessage.append("\t\t --gpu-sm run GPU version of algorithm with shared memory optimization\n");
119 |     helpMessage.append("\t\t --gpu-mono run GPU version of algorithm with single kernel arch on single block\n");
120 |     helpMessage.append("\t\t --run-all run all implemented versions of Otsu algorithm (CPU and GPU)\n");
121 | 
122 |     printf(helpMessage.c_str());
123 | }
124 | 
125 | int AppRunner::parseIntInputParam(const char *param, int defaultValue)
126 | {
127 |     return std::atoi(param) > 0 ? std::atoi(param) : defaultValue;
128 | }


--------------------------------------------------------------------------------
/src/main.cpp:
--------------------------------------------------------------------------------
  1 | // System includes
  2 | #include <stdio.h>
  3 | #include <stdlib.h>
  4 | #include <assert.h>
  5 | #include <iostream>
  6 | #include <time.h>
  7 | #include <omp.h>
  8 | 
  9 | // Memory leaks checking
 10 | // #include "vld.h"
 11 | 
 12 | #include "libs/lodepng.h"
 13 | #include "core/AppRunner.h"
 14 | #include "utils/ImageFileUtil.h"
 15 | #include "utils/CudaUtil.h"
 16 | #include "utils/RunConfigurationBuilder.h"
 17 | #include "model/PngImage.h"
 18 | #include "model/RunConfiguration.h"
 19 | #include "core/binarizers/OtsuBinarizer.h"
 20 | #include "core/binarizers/OtsuOpenMPBinarizer.h"
 21 | #include "core/binarizers/CudaOtsuBinarizer.cuh"
 22 | #include "core/binarizers/SMCudaOtsuBinarizer.cuh"
 23 | #include "core/binarizers/MonoCudaOtsuBinarizer.cuh"
 24 | 
 25 | std::string getConfigurationInfo(int threadsPerBlock, int numBlocks)
 26 | {
 27 | 	std::vector<std::string> params;
 28 | 	params.push_back(std::to_string(threadsPerBlock));
 29 | 	params.push_back(std::to_string(numBlocks));
 30 | 	params.push_back(CudaUtil::getDeviceName(CudaUtil::getCurrentDevice()));
 31 | 
 32 | 	return ImageFileUtil::joinString(params, ',');
 33 | }
 34 | 
 35 | std::string runGpuImplementation(RunConfiguration* runConfig)
 36 | {
 37 | 
 38 | 	CudaOtsuBinarizer *cudaBinarizer = new CudaOtsuBinarizer(
 39 | 		runConfig->getThreadsPerBlock(),
 40 | 		runConfig->getNumberOfBlocks(),
 41 | 		runConfig->shouldDrawHistograms());
 42 | 
 43 | 	PngImage *gpuBinarizedImage = cudaBinarizer->binarize(runConfig->getLoadedImage());
 44 | 
 45 | 	std::string gpuBinarizedFilename = ImageFileUtil::addPrefix(runConfig->getFullFilePath(), "gpu_binarized_");
 46 | 
 47 | 	ImageFileUtil::savePngFile(gpuBinarizedImage, gpuBinarizedFilename.c_str());
 48 | 
 49 | 	std::string csvTimesLog = cudaBinarizer->getBinarizerExecutionInfo(runConfig->getFullFilePath());
 50 | 	std::string configLog = getConfigurationInfo(
 51 | 		runConfig->getThreadsPerBlock(),
 52 | 		runConfig->getNumberOfBlocks());
 53 | 
 54 | 	delete gpuBinarizedImage;
 55 | 	delete cudaBinarizer;
 56 | 
 57 | 	return csvTimesLog + "," + configLog;
 58 | }
 59 | 
 60 | std::string runGpuSharedMemoryImplementation(RunConfiguration* runConfig)
 61 | {
 62 | 
 63 | 	SMCudaOtsuBinarizer *smCudaBinarizer = new SMCudaOtsuBinarizer(
 64 | 		runConfig->getThreadsPerBlock(),
 65 | 		runConfig->getNumberOfBlocks(),
 66 | 		runConfig->shouldDrawHistograms());
 67 | 
 68 | 	PngImage *sharedMemoryGpuBinarizedImage = smCudaBinarizer->binarize(runConfig->getLoadedImage());
 69 | 
 70 | 	std::string smGpuBinarizedFilename = ImageFileUtil::addPrefix(runConfig->getFullFilePath(), "gpu_shared_memory_binarized_");
 71 | 
 72 | 	ImageFileUtil::savePngFile(sharedMemoryGpuBinarizedImage, smGpuBinarizedFilename.c_str());
 73 | 
 74 | 	std::string csvTimesLog = smCudaBinarizer->getBinarizerExecutionInfo(runConfig->getFullFilePath());
 75 | 	std::string configLog = getConfigurationInfo(runConfig->getThreadsPerBlock(), runConfig->getNumberOfBlocks());
 76 | 
 77 | 	delete sharedMemoryGpuBinarizedImage;
 78 | 	delete smCudaBinarizer;
 79 | 
 80 | 	return csvTimesLog + "," + configLog;
 81 | }
 82 | 
 83 | void runGpuMonoKernelImplementation(RunConfiguration* runConfig)
 84 | {
 85 | 
 86 | 	MonoCudaOtsuBinarizer *monoCudaBinarizer = new MonoCudaOtsuBinarizer(runConfig->getThreadsPerBlock(), runConfig->shouldDrawHistograms());
 87 | 
 88 | 	PngImage *monoKernelGpuBinarizedImage = monoCudaBinarizer->binarize(runConfig->getLoadedImage());
 89 | 
 90 | 	std::string monoKernelGpuBinarizedFilename = ImageFileUtil::addPrefix(runConfig->getFullFilePath(), "gpu_mono_binarized_");
 91 | 
 92 | 	ImageFileUtil::savePngFile(monoKernelGpuBinarizedImage, monoKernelGpuBinarizedFilename.c_str());
 93 | 
 94 | 	delete monoKernelGpuBinarizedImage;
 95 | 	delete monoCudaBinarizer;
 96 | }
 97 | 
 98 | // CudaOtsu filepath/dirpath threadsPerBlock numBlocks
 99 | int main(int argc, char **argv)
100 | {
101 | 	std::vector<std::string> binarizerTimestamps;
102 | 	const char *timestampsFile = "times.csv";
103 | 
104 | 	AppRunner *appRunner = new AppRunner();
105 | 
106 | 	appRunner->loadInputConfiguration(argc, argv);
107 | 
108 | 	RunConfiguration* runConfig = appRunner->getRunConfig();
109 | 
110 | 	runConfig->print();
111 | 
112 | 	if (runConfig->hasLoadedImage())
113 | 	{
114 | 		// To refactor
115 | 
116 | 		if (runConfig->shouldRunAlgorithm(CPU))
117 | 		{
118 | 			OtsuBinarizer* cpuBinarizer = new OtsuBinarizer(runConfig->getLoadedImage());
119 | 			cpuBinarizer->binarize(runConfig)->printResult();
120 | 		}
121 | 
122 | 		if (runConfig->shouldRunAlgorithm(CPU_OpenMP))
123 | 		{
124 | 			OtsuOpenMPBinarizer* openMpBinarizer = new OtsuOpenMPBinarizer(runConfig->getLoadedImage());
125 | 			openMpBinarizer->binarize(runConfig)->printResult();
126 | 		}
127 | 
128 | 		if (runConfig->shouldRunAlgorithm(GPU))
129 | 		{
130 | 			std::string csvTimeLog = runGpuImplementation(runConfig);
131 | 			binarizerTimestamps.push_back(csvTimeLog);
132 | 		}
133 | 
134 | 		if (runConfig->shouldRunAlgorithm(GPU_SharedMemory))
135 | 		{
136 | 			std::string csvTimeLog = runGpuSharedMemoryImplementation(runConfig);
137 | 			binarizerTimestamps.push_back(csvTimeLog);
138 | 		}
139 | 
140 | 		if (runConfig->shouldRunAlgorithm(GPU_MonoKernel))
141 | 		{
142 | 			runGpuMonoKernelImplementation(runConfig);
143 | 		}
144 | 	} else {
145 | 		printf("\nFile not loaded");
146 | 	}
147 | 
148 | 	ImageFileUtil::saveCsvFile(binarizerTimestamps, timestampsFile);
149 | 
150 | 	return 0;
151 | }
152 | 


--------------------------------------------------------------------------------
/src/core/binarizers/OtsuOpenMPBinarizer.cpp:
--------------------------------------------------------------------------------
  1 | #include "../../utils/ImageFileUtil.h"
  2 | #include "OtsuOpenMPBinarizer.h"
  3 | #include <omp.h>
  4 | #include <stdio.h>
  5 | #include <time.h>
  6 | 
  7 | OtsuOpenMPBinarizer::OtsuOpenMPBinarizer(PngImage* imageToBinarize)
  8 | {
  9 | 	this->imageToBinarize = imageToBinarize;
 10 | }
 11 | 
 12 | MethodImplementation OtsuOpenMPBinarizer::getBinarizerType()
 13 | {
 14 | 	return CPU_OpenMP;
 15 | }
 16 | 
 17 | const char* OtsuOpenMPBinarizer::getBinarizedFilePrefix()
 18 | {
 19 | 	return "cpu-openmp_binarized_";
 20 | }
 21 | 
 22 | BinarizationResult* OtsuOpenMPBinarizer::binarize(RunConfiguration* runConfig)
 23 | {
 24 | 	printf("\nSetting OpenMP threads num to %d threads\n", runConfig->getCpuThreads());
 25 | 	omp_set_dynamic(0);
 26 | 	omp_set_num_threads(runConfig->getCpuThreads());
 27 | 
 28 | 	this->histogram = calculateHistogram();
 29 | 	this->cpuThreads = runConfig->getCpuThreads();
 30 | 
 31 | 	ExecutionTimestamp* executionTimestamp = new ExecutionTimestamp();
 32 | 	clock_t time;
 33 | 	time = clock();
 34 | 
 35 | 	std::string cpuBinarizedFilename = ImageFileUtil::addPrefix(runConfig->getFullFilePath(), getBinarizedFilePrefix());
 36 | 
 37 | 	PngImage *cpuBinarizedImage = binarize();
 38 | 
 39 | 	time = clock() - time;
 40 | 
 41 | 	executionTimestamp->binarizationTimeInSeconds = ((double)time / CLOCKS_PER_SEC);
 42 | 
 43 | 	printf("\nCPU-OpenMP binarization taken %f seconds\n", ((double)time / CLOCKS_PER_SEC));
 44 | 
 45 | 	ImageFileUtil::savePngFile(cpuBinarizedImage, cpuBinarizedFilename.c_str());
 46 | 
 47 | 	delete cpuBinarizedImage;
 48 | 
 49 | 	return new BinarizationResult(
 50 | 		getBinarizerType(),
 51 | 		cpuBinarizedFilename.c_str(),
 52 | 		executionTimestamp
 53 | 	);
 54 | }
 55 | 
 56 | std::vector<double> OtsuOpenMPBinarizer::calculateHistogram() {
 57 | 	std::vector<double> histogram(PngImage::MAX_PIXEL_VALUE);
 58 | 	std::vector<unsigned char> image = this->imageToBinarize->getRawPixelData();
 59 | 
 60 | 	unsigned char pixelValue;
 61 | 	long totalPixels = image.size();
 62 | 
 63 | 	#pragma omp parallel firstprivate(pixelValue) shared(totalPixels, histogram, image) num_threads(cpuThreads)
 64 | 	{
 65 | 		int chunkSize = PngImage::MAX_PIXEL_VALUE / omp_get_num_threads();
 66 | 
 67 | 		#pragma omp for schedule(static, chunkSize)
 68 | 		for (int i = 0; i < totalPixels; i++) {
 69 | 			pixelValue = image[i];
 70 | 			
 71 | 			#pragma omp atomic
 72 | 			histogram[pixelValue]++;
 73 | 		}
 74 | 
 75 | 		#pragma omp barrier
 76 | 
 77 | 		// Normalization
 78 | 		#pragma omp for schedule(static, chunkSize)
 79 | 		for (int v = 0; v < PngImage::MAX_PIXEL_VALUE; v++) {
 80 | 			histogram[v] /= totalPixels;
 81 | 		}
 82 | 	}
 83 | 
 84 | 	return histogram;
 85 | }
 86 | 
 87 | int OtsuOpenMPBinarizer::findThreshold() {
 88 | 	int threshold;
 89 | 	long int totalPixels = this->imageToBinarize->getTotalPixels();
 90 | 	double* betweenClassVariances = new double[PngImage::MAX_PIXEL_VALUE];
 91 | 	double allProbabilitySum = 0;
 92 | 	std::vector<double> localHistogram = this->histogram;
 93 | 
 94 | 	#pragma omp parallel shared(allProbabilitySum, betweenClassVariances, totalPixels, localHistogram) num_threads(cpuThreads)
 95 | 	{
 96 | 		double firstClassProbability = 0, secondClassProbability = 0;
 97 | 		double firstClassMean = 0, secondClassMean = 0, firstProbabilitySum = 0;
 98 | 
 99 | 		int chunkSize = PngImage::MAX_PIXEL_VALUE / omp_get_num_threads();
100 | 
101 | 		#pragma omp for schedule(static, chunkSize)
102 | 		for (int i = 0; i < PngImage::MAX_PIXEL_VALUE; i++) {
103 | 			#pragma omp atomic
104 | 			allProbabilitySum += i * localHistogram[i];
105 | 			betweenClassVariances[i] = 0;
106 | 		}
107 | 
108 | 		#pragma omp barrier
109 | 
110 | 		#pragma omp for schedule(static, chunkSize)
111 | 		for (int v = 0; v < PngImage::MAX_PIXEL_VALUE; v++) {
112 | 			firstClassProbability = 0;
113 | 			firstProbabilitySum = 0;
114 | 			for (int t = 0; t <= v % PngImage::MAX_PIXEL_VALUE; t++) {
115 | 				firstClassProbability += localHistogram[t];
116 | 				firstProbabilitySum += t * localHistogram[t];
117 | 			}
118 | 
119 | 			secondClassProbability = 1 - firstClassProbability;
120 | 
121 | 			firstClassMean = (double)firstProbabilitySum / (double)firstClassProbability;
122 | 			secondClassMean = (double)(allProbabilitySum - firstProbabilitySum) / (double)secondClassProbability;
123 | 
124 | 			betweenClassVariances[v] = firstClassProbability * secondClassProbability * pow((firstClassMean - secondClassMean), 2);
125 | 		}
126 | 
127 | 		#pragma omp barrier
128 | 
129 | 		#pragma omp single 
130 | 		{
131 | 			double maxVariance = 0;
132 | 			for (int v = 0; v < PngImage::MAX_PIXEL_VALUE; v++) {
133 | 				if (betweenClassVariances[v] > maxVariance) {
134 | 					threshold = v;
135 | 					maxVariance = betweenClassVariances[v];
136 | 				}
137 | 			}
138 | 		}
139 | 	}
140 | 
141 | 	delete[] betweenClassVariances;
142 | 
143 | 	return threshold;
144 | }
145 | 
146 | PngImage* OtsuOpenMPBinarizer::binarize() {
147 | 	int threshold = findThreshold();
148 | 	std::vector<unsigned char> imagePixels = imageToBinarize->getRawPixelData();
149 | 
150 | 	int totalPixels = imageToBinarize->getTotalPixels();
151 | 	int chunkSize = totalPixels / cpuThreads;
152 | 
153 | 	#pragma omp parallel for shared(totalPixels, imagePixels, threshold) schedule(dynamic, chunkSize) num_threads(cpuThreads)
154 | 	for (int i = 0; i < totalPixels; i++) {
155 | 		if ((int)imagePixels[i] > threshold) {
156 | 			imagePixels[i] = PngImage::COLOR_WHITE;
157 | 		}
158 | 		else {
159 | 			imagePixels[i] = PngImage::COLOR_BLACK;
160 | 		}
161 | 	}
162 | 
163 | 	return new PngImage(
164 | 		imageToBinarize->getFilename(),
165 | 		imageToBinarize->getWidth(),
166 | 		imageToBinarize->getHeight(),
167 | 		imagePixels
168 | 	);
169 | }
170 | 
171 | void OtsuOpenMPBinarizer::showHistogram() {
172 | 	printf("\nHistogram:\n");
173 | 	double value = 0;
174 | 	for (int i = 0; i < PngImage::MAX_PIXEL_VALUE; i++) {
175 | 		value = histogram[i];
176 | 		printf("\tPixel value %d -> %.5f\n", i, value);
177 | 	}
178 | }
179 | 


--------------------------------------------------------------------------------
/src/core/binarizers/SMCudaOtsuBinarizer.cu:
--------------------------------------------------------------------------------
  1 | #include "SMCudaOtsuBinarizer.cuh"
  2 | 
  3 | #include <stdio.h>
  4 | 
  5 | // CUDA imports
  6 | #include <cuda_runtime.h>
  7 | 
  8 | __shared__ unsigned int myHistogram[PngImage::MAX_PIXEL_VALUE];
  9 | __global__ void smKernelCalculateHistogram(unsigned int* histogram, unsigned char* rawPixels, long chunkSize, long totalPixels, int histogramChunk)
 10 | {
 11 | 	int id = blockDim.x * blockIdx.x + threadIdx.x;
 12 | 	
 13 | 	if (histogramChunk > 1) {
 14 | 		int histogramStartPosition = threadIdx.x * histogramChunk;
 15 | 		for (int h = histogramStartPosition; h < (histogramStartPosition + histogramChunk); h++) {
 16 | 			if (h < PngImage::MAX_PIXEL_VALUE) {
 17 | 				myHistogram[h] = 0;
 18 | 			}
 19 | 		}
 20 | 	} else {
 21 | 		if (threadIdx.x < PngImage::MAX_PIXEL_VALUE) { 
 22 | 			myHistogram[threadIdx.x] = 0;
 23 | 		}
 24 | 	}
 25 | 
 26 | 	__syncthreads();
 27 | 
 28 | 	int startPosition = id * chunkSize;
 29 | 	for (int i = startPosition; i < (startPosition + chunkSize); i++) {
 30 | 		if (i < totalPixels) {
 31 | 			int pixelValue = (int)rawPixels[i];
 32 | 			atomicAdd(&myHistogram[pixelValue], 1);
 33 | 		}
 34 | 	}
 35 | 
 36 | 	__syncthreads();
 37 | 
 38 | 	if (histogramChunk > 1) {
 39 | 		int histogramStartPosition = threadIdx.x * histogramChunk;
 40 | 		for (int h = histogramStartPosition; h < (histogramStartPosition + histogramChunk); h++) {
 41 | 			if (h < PngImage::MAX_PIXEL_VALUE) {
 42 | 				atomicAdd(&histogram[h], myHistogram[h]);
 43 | 			}
 44 | 		}
 45 | 	} else {
 46 | 		if (threadIdx.x < PngImage::MAX_PIXEL_VALUE) {
 47 | 			atomicAdd(&histogram[threadIdx.x], myHistogram[threadIdx.x]);
 48 | 		}
 49 | 	}
 50 | }
 51 | 
 52 | __shared__ double myHistogramCopy[PngImage::MAX_PIXEL_VALUE];
 53 | __global__ void smKernelComputeClassVariances(double* histogram, double allProbabilitySum, long int totalPixels, double* betweenClassVariance)
 54 | {
 55 | 	int id = blockDim.x * blockIdx.x + threadIdx.x;
 56 | 
 57 | 	myHistogramCopy[threadIdx.x % PngImage::MAX_PIXEL_VALUE] = histogram[threadIdx.x % PngImage::MAX_PIXEL_VALUE];
 58 | 	
 59 | 	__syncthreads();
 60 | 
 61 | 	double firstClassProbability = 0, secondClassProbability = 0;
 62 | 	double firstClassMean = 0, secondClassMean = 0;
 63 | 	double firstProbabilitySum = 0;
 64 | 
 65 | 	for (int t = 0; t <= id % PngImage::MAX_PIXEL_VALUE; t++) {
 66 | 		firstClassProbability += myHistogramCopy[t];
 67 | 		firstProbabilitySum += t * myHistogramCopy[t];
 68 | 	}
 69 | 
 70 | 	secondClassProbability = 1 - firstClassProbability;
 71 | 
 72 | 	firstClassMean = (double)firstProbabilitySum / (double)firstClassProbability;
 73 | 	secondClassMean = (double)(allProbabilitySum - firstProbabilitySum) / (double)secondClassProbability;
 74 | 
 75 | 	betweenClassVariance[id] = firstClassProbability * secondClassProbability * pow((firstClassMean - secondClassMean), 2);
 76 | }
 77 | 
 78 | SMCudaOtsuBinarizer::SMCudaOtsuBinarizer(int threadsPerBlock, int numBlocks, bool drawHistogram) : CudaOtsuBinarizer(threadsPerBlock, numBlocks, drawHistogram, "GPU - Shared Memory") {
 79 | 	this->threadsPerBlock_ = threadsPerBlock;
 80 | 	this->numBlocks_ = numBlocks;
 81 | 	this->drawHistogram_ = drawHistogram;
 82 | }
 83 | 
 84 | SMCudaOtsuBinarizer::~SMCudaOtsuBinarizer() {}
 85 | 
 86 | unsigned char SMCudaOtsuBinarizer::cudaFindThreshold(double* histogram, long int totalPixels) {
 87 | 	cudaEvent_t start, stop;
 88 | 	cudaEventCreate(&start);
 89 | 	cudaEventCreate(&stop);
 90 | 	int threadsPerBlock = 256;
 91 | 	int numBlocks = 1;
 92 | 
 93 | 	double allProbabilitySum = 0;
 94 | 	for (int i = 0; i < PngImage::MAX_PIXEL_VALUE; i++) {
 95 | 		allProbabilitySum += i * histogram[i];
 96 | 	}
 97 | 
 98 | 	double* hostBetweenClassVariances = new double[PngImage::MAX_PIXEL_VALUE];
 99 | 	for (int i = 0; i < PngImage::MAX_PIXEL_VALUE; i++) {
100 | 		hostBetweenClassVariances[i] = 0;
101 | 	}
102 | 
103 | 	double* deviceHistogram;
104 | 	cudaMalloc((void **)&deviceHistogram, sizeof(double) * PngImage::MAX_PIXEL_VALUE);
105 | 	cudaMemcpy(deviceHistogram, histogram, sizeof(double) * PngImage::MAX_PIXEL_VALUE, cudaMemcpyHostToDevice);
106 | 
107 | 	double* deviceBetweenClassVariances;
108 | 	cudaMalloc((void **)&deviceBetweenClassVariances, sizeof(double) * PngImage::MAX_PIXEL_VALUE);
109 | 	cudaMemcpy(deviceBetweenClassVariances, hostBetweenClassVariances, sizeof(double) * PngImage::MAX_PIXEL_VALUE, cudaMemcpyHostToDevice);
110 | 
111 | 	cudaEventRecord(start);
112 | 	smKernelComputeClassVariances << <numBlocks, threadsPerBlock >> >(deviceHistogram, allProbabilitySum, totalPixels, deviceBetweenClassVariances);
113 | 	cudaEventRecord(stop);
114 | 	cudaMemcpy(hostBetweenClassVariances, deviceBetweenClassVariances, sizeof(double) * PngImage::MAX_PIXEL_VALUE, cudaMemcpyDeviceToHost);
115 | 
116 | 	cudaEventSynchronize(stop);
117 | 	float milliseconds = 0;
118 | 	cudaEventElapsedTime(&milliseconds, start, stop);
119 | 	printf("\n\t[%s] Threshold calculated in %.6f milliseconds \n", this->TAG, milliseconds);
120 | 	binarizerTimestamp_->thresholdFindingTimeInSeconds += milliseconds / 1000;
121 | 
122 | 	cudaFree(deviceHistogram);
123 | 	cudaFree(deviceBetweenClassVariances);
124 | 
125 | 	double maxVariance = 0;
126 | 	unsigned char currentBestThreshold = 0;
127 | 	for (int t = 0; t < PngImage::MAX_PIXEL_VALUE; t++) {
128 | 		if (hostBetweenClassVariances[t] > maxVariance) {
129 | 			currentBestThreshold = (unsigned char)t;
130 | 			maxVariance = hostBetweenClassVariances[t];
131 | 		}
132 | 	}
133 | 
134 | 	delete hostBetweenClassVariances;
135 | 
136 | 	return currentBestThreshold;
137 | }
138 | 
139 | double* SMCudaOtsuBinarizer::cudaCalculateHistogram(unsigned char* rawPixels, long totalPixels) {
140 | 	//TODO: check cudaGetDeviceProperties function!
141 | 
142 | 	cudaEvent_t start, stop;
143 | 	cudaEventCreate(&start);
144 | 	cudaEventCreate(&stop);
145 | 
146 | 	unsigned int* hostHistogram = new unsigned int[PngImage::MAX_PIXEL_VALUE];
147 | 	for (int i = 0; i < PngImage::MAX_PIXEL_VALUE; i++) {
148 | 		hostHistogram[i] = 0;
149 | 	}
150 | 
151 | 	unsigned int* deviceHistogram;
152 | 	cudaMalloc((void **)&deviceHistogram, sizeof(unsigned int) * PngImage::MAX_PIXEL_VALUE);
153 | 	cudaMemcpy(deviceHistogram, hostHistogram, sizeof(unsigned int) * PngImage::MAX_PIXEL_VALUE, cudaMemcpyHostToDevice);
154 | 
155 | 	unsigned char* deviceRawPixels;
156 | 	cudaMalloc((void **)&deviceRawPixels, sizeof(unsigned char) * totalPixels);
157 | 	cudaMemcpy(deviceRawPixels, rawPixels, sizeof(unsigned char) * totalPixels, cudaMemcpyHostToDevice);
158 | 
159 | 	long chunkSize = ceil(totalPixels / (threadsPerBlock_ * numBlocks_)) + 1;
160 | 
161 | 	int histogramChunk = ceil(PngImage::MAX_PIXEL_VALUE / threadsPerBlock_) + 1;
162 | 
163 | 	cudaEventRecord(start);
164 | 	smKernelCalculateHistogram<<<numBlocks_, threadsPerBlock_ >>>(deviceHistogram, deviceRawPixels, chunkSize, totalPixels, histogramChunk);
165 | 	cudaEventRecord(stop);
166 | 
167 | 	cudaMemcpy(hostHistogram, deviceHistogram, sizeof(unsigned int) * PngImage::MAX_PIXEL_VALUE, cudaMemcpyDeviceToHost);
168 | 
169 | 	cudaEventSynchronize(stop);
170 | 	float milliseconds = 0;
171 | 	cudaEventElapsedTime(&milliseconds, start, stop);
172 | 	printf("\n\t[%s] Histogram calculated in %.6f milliseconds \n", this->TAG, milliseconds);
173 | 	binarizerTimestamp_->histogramBuildingTimeInSeconds += milliseconds / 1000;
174 | 
175 | 	cudaFree(deviceHistogram);
176 | 	cudaFree(deviceRawPixels);
177 | 
178 | 	double* normalizedHistogram = new double[PngImage::MAX_PIXEL_VALUE];
179 | 	long pixelsSum = 0;
180 | 	for (int v = 0; v < PngImage::MAX_PIXEL_VALUE; v++) {
181 | 		normalizedHistogram[v] = (double)hostHistogram[v] / (double)totalPixels;
182 | 		pixelsSum += hostHistogram[v];
183 | 	}
184 | 	printf("\n\t[%s] Histogram pixels: %d \n", this->TAG, pixelsSum);
185 | 
186 | 	delete hostHistogram;
187 | 
188 | 	return normalizedHistogram;
189 | }
190 | 


--------------------------------------------------------------------------------
/src/core/binarizers/MonoCudaOtsuBinarizer.cu:
--------------------------------------------------------------------------------
  1 | #include "MonoCudaOtsuBinarizer.cuh"
  2 | 
  3 | #include <stdio.h>
  4 | 
  5 | // CUDA imports
  6 | #include <cuda_runtime.h>
  7 | 
  8 | __global__ void kernelBinarize(unsigned int* histogram, unsigned char* rawPixels, double* betweenClassVariances, double *allProbabilitySum,
  9 | 	unsigned int* threshold, long totalPixels, int threadsPerBlock)
 10 | {
 11 | 	int id = blockDim.x * blockIdx.x + threadIdx.x;
 12 | 
 13 | 	int chunkSize = ceil((float)totalPixels / (float)(threadsPerBlock));
 14 | 	int startPosition = id * chunkSize;
 15 | 	
 16 | 	// Calculate Histogram
 17 | 	for (int i = startPosition; i < (startPosition + chunkSize); i++) {
 18 | 		if (i < totalPixels) {
 19 | 			int pixelValue = (int)rawPixels[i];
 20 | 			atomicAdd(&histogram[pixelValue], 1);
 21 | 		}
 22 | 	}
 23 | 
 24 | 	__syncthreads();
 25 | 
 26 | 	// Compute best class variance
 27 | 
 28 | 	if (id == 0) {
 29 | 		for (int i = 0; i < PngImage::MAX_PIXEL_VALUE; i++) {
 30 | 			*allProbabilitySum += i * ((double)histogram[i] / (double)totalPixels);
 31 | 			betweenClassVariances[i] = 0;
 32 | 		}
 33 | 	}
 34 | 
 35 | 	__syncthreads();
 36 | 
 37 | 	int histogramChunk = ceil((float)PngImage::MAX_PIXEL_VALUE / (float)(threadsPerBlock));
 38 | 
 39 | 	if (id < PngImage::MAX_PIXEL_VALUE) {
 40 | 		double firstClassProbability = 0, secondClassProbability = 0;
 41 | 		double firstClassMean = 0, secondClassMean = 0;
 42 | 		double firstProbabilitySum = 0;
 43 | 
 44 | 		int histogramStartPosition = id * histogramChunk;
 45 | 		double normalizedHistogramValue;
 46 | 		for (int h = histogramStartPosition; h < (histogramStartPosition + histogramChunk); h++) {
 47 | 			if (h < PngImage::MAX_PIXEL_VALUE) {
 48 | 				firstClassProbability = 0;
 49 | 				firstProbabilitySum = 0;
 50 | 				for (int t = 0; t <= h; t++) {
 51 | 					normalizedHistogramValue = ((double)histogram[t] / (double)totalPixels);
 52 | 					firstClassProbability += normalizedHistogramValue;
 53 | 					firstProbabilitySum += t * normalizedHistogramValue;
 54 | 				}
 55 | 
 56 | 				secondClassProbability = 1 - firstClassProbability;
 57 | 
 58 | 				firstClassMean = (double)firstProbabilitySum / (double)firstClassProbability;
 59 | 				secondClassMean = (double)(*allProbabilitySum - firstProbabilitySum) / (double)secondClassProbability;
 60 | 
 61 | 				betweenClassVariances[h] = firstClassProbability * secondClassProbability * pow((firstClassMean - secondClassMean), 2);
 62 | 			}
 63 | 		}
 64 | 	}
 65 | 
 66 | 	__syncthreads();
 67 | 
 68 | 	if (id == 0) {
 69 | 		double maxVariance = 0;
 70 | 		unsigned int currentBestThreshold = 0;
 71 | 		for (int t = 0; t < PngImage::MAX_PIXEL_VALUE; t++) {
 72 | 			if (betweenClassVariances[t] > maxVariance) {
 73 | 				currentBestThreshold = t;
 74 | 				maxVariance = betweenClassVariances[t];
 75 | 			}
 76 | 		}
 77 | 
 78 | 		*threshold = currentBestThreshold;
 79 | 	}
 80 | 
 81 | 	__syncthreads();
 82 | 
 83 | 	int bestThreshold = *threshold;
 84 | 	for (int i = startPosition; i < (startPosition + chunkSize); i++) {
 85 | 		if (i < totalPixels) {
 86 | 			if ((int)rawPixels[i] > bestThreshold) {
 87 | 				rawPixels[i] = PngImage::COLOR_WHITE;
 88 | 			}
 89 | 			else {
 90 | 				rawPixels[i] = PngImage::COLOR_BLACK;
 91 | 			}
 92 | 		}
 93 | 	}
 94 | }
 95 | 
 96 | MonoCudaOtsuBinarizer::MonoCudaOtsuBinarizer(int threadsPerBlock, bool drawHistogram, const char* TAG) {
 97 | 	this->threadsPerBlock_ = threadsPerBlock;
 98 | 	this->executionTime_ = 0;
 99 | 
100 | 	this->showHistogram_ = drawHistogram;
101 | 	this->TAG = TAG;
102 | }
103 | 
104 | MonoCudaOtsuBinarizer::~MonoCudaOtsuBinarizer() {}
105 | 
106 | PngImage* MonoCudaOtsuBinarizer::binarize(PngImage * imageToBinarize)
107 | {
108 | 	long totalImagePixels = (long)imageToBinarize->getRawPixelData().size();
109 | 
110 | 	unsigned char* binarizedRawPixels = cudaBinarize(imageToBinarize->getRawPixelData().data(), totalImagePixels);
111 | 	cudaDeviceSynchronize();
112 | 
113 | 	std::vector<unsigned char> binarizedVector(&binarizedRawPixels[0], &binarizedRawPixels[totalImagePixels]);
114 | 
115 | 	delete binarizedRawPixels;
116 | 
117 | 	printf("\n\t[%s] Total calculation time: %.6f milliseconds \n", this->TAG, this->executionTime_);
118 | 
119 | 	return new PngImage(
120 | 		imageToBinarize->getFilename(),
121 | 		imageToBinarize->getWidth(),
122 | 		imageToBinarize->getHeight(),
123 | 		binarizedVector
124 | 	);
125 | }
126 | 
127 | void MonoCudaOtsuBinarizer::showHistogram(double* histogram) {
128 | 	printf("\nHistogram:\n");
129 | 	double value = 0;
130 | 	for (int i = 0; i < PngImage::MAX_PIXEL_VALUE; i++) {
131 | 		value = histogram[i];
132 | 		printf("\tPixel value %d -> %.5f\n", i, value);
133 | 	}
134 | }
135 | 
136 | unsigned char* MonoCudaOtsuBinarizer::cudaBinarize(unsigned char * rawPixels, long totalPixels) {
137 | 	cudaEvent_t start, stop;
138 | 	cudaEventCreate(&start);
139 | 	cudaEventCreate(&stop);
140 | 
141 | 	unsigned char* hostRawPixels = new unsigned char[totalPixels];
142 | 
143 | 	unsigned char* deviceRawPixels;
144 | 	cudaMalloc((void **)&deviceRawPixels, sizeof(unsigned char) * totalPixels);
145 | 	cudaMemcpy(deviceRawPixels, rawPixels, totalPixels * sizeof(unsigned char), cudaMemcpyHostToDevice);
146 | 
147 | 	unsigned int* hostHistogram = new unsigned int[PngImage::MAX_PIXEL_VALUE];
148 | 	for (int i = 0; i < PngImage::MAX_PIXEL_VALUE; i++) {
149 | 		hostHistogram[i] = 0;
150 | 	}
151 | 
152 | 	double* hostBetweenClassVariances = new double[PngImage::MAX_PIXEL_VALUE];
153 | 	for (int i = 0; i < PngImage::MAX_PIXEL_VALUE; i++) {
154 | 		hostBetweenClassVariances[i] = 0;
155 | 	}
156 | 
157 | 	double* hostAllProbabilitySum = new double;
158 | 	*hostAllProbabilitySum = 0;
159 | 
160 | 	double* deviceAllProbabilitySum;
161 | 	cudaMalloc((void **)&deviceAllProbabilitySum, sizeof(double));
162 | 	cudaMemcpy(deviceAllProbabilitySum, hostAllProbabilitySum, sizeof(double), cudaMemcpyHostToDevice);
163 | 
164 | 	unsigned int hostThreshold = 0;
165 | 
166 | 	unsigned int* deviceThreshold;
167 | 	cudaMalloc((void **)&deviceThreshold, sizeof(unsigned int));
168 | 	cudaMemcpy(deviceThreshold, &hostThreshold, sizeof(unsigned int), cudaMemcpyHostToDevice);
169 | 
170 | 	unsigned int* deviceHistogram;
171 | 	cudaMalloc((void **)&deviceHistogram, sizeof(unsigned int) * PngImage::MAX_PIXEL_VALUE);
172 | 	cudaMemcpy(deviceHistogram, hostHistogram, sizeof(unsigned int) * PngImage::MAX_PIXEL_VALUE, cudaMemcpyHostToDevice);
173 | 
174 | 	double* deviceBetweenClassVariances;
175 | 	cudaMalloc((void **)&deviceBetweenClassVariances, sizeof(double) * PngImage::MAX_PIXEL_VALUE);
176 | 	cudaMemcpy(deviceBetweenClassVariances, hostBetweenClassVariances, sizeof(double) * PngImage::MAX_PIXEL_VALUE, cudaMemcpyHostToDevice);
177 | 
178 | 	cudaEventRecord(start);
179 | 	kernelBinarize<<<1, threadsPerBlock_ >>>(deviceHistogram, deviceRawPixels, deviceBetweenClassVariances, deviceAllProbabilitySum,
180 | 		deviceThreshold, totalPixels, threadsPerBlock_);
181 | 	cudaEventRecord(stop);
182 | 
183 | 	cudaMemcpy(hostHistogram, deviceHistogram, sizeof(unsigned int) * PngImage::MAX_PIXEL_VALUE, cudaMemcpyDeviceToHost);
184 | 
185 | 	cudaDeviceSynchronize();
186 | 
187 | 	
188 | 	double* normalizedHistogram = new double[PngImage::MAX_PIXEL_VALUE];
189 | 	long pixelsSum = 0;
190 | 	for (int v = 0; v < PngImage::MAX_PIXEL_VALUE; v++) {
191 | 		normalizedHistogram[v] = (double)hostHistogram[v] / (double)totalPixels;
192 | 		pixelsSum += hostHistogram[v];
193 | 	}
194 | 
195 | 	if (this->showHistogram_) {
196 | 		printf("\n\t[%s] Histogram pixels: %d \n", this->TAG, pixelsSum);
197 | 		showHistogram(normalizedHistogram);
198 | 	}
199 | 	
200 | 	cudaFree(deviceHistogram);
201 | 	cudaFree(deviceBetweenClassVariances);
202 | 	cudaFree(deviceAllProbabilitySum);
203 | 	cudaFree(deviceThreshold);
204 | 
205 | 	cudaMemcpy(hostRawPixels, deviceRawPixels, sizeof(unsigned char) * totalPixels, cudaMemcpyDeviceToHost);
206 | 
207 | 	cudaEventSynchronize(stop);
208 | 	float milliseconds = 0;
209 | 	cudaEventElapsedTime(&milliseconds, start, stop);
210 | 	this->executionTime_ += milliseconds;
211 | 
212 | 	cudaFree(deviceRawPixels);
213 | 
214 | 	delete hostHistogram;
215 | 	delete hostBetweenClassVariances;
216 | 	delete hostAllProbabilitySum;
217 | 	delete normalizedHistogram;
218 | 
219 | 	return hostRawPixels;
220 | }
221 | 


--------------------------------------------------------------------------------
/src/core/binarizers/CudaOtsuBinarizer.cu:
--------------------------------------------------------------------------------
  1 | #include "CudaOtsuBinarizer.cuh"
  2 | 
  3 | #include <stdio.h>
  4 | 
  5 | // CUDA imports
  6 | #include <cuda_runtime.h>
  7 | 
  8 | __global__ void kernelCalculateHistogram(unsigned int* histogram, unsigned char* rawPixels, long chunkSize, long totalPixels)
  9 | {
 10 | 	int id = blockDim.x * blockIdx.x + threadIdx.x;
 11 | 
 12 | 	int startPosition = id * chunkSize;
 13 | 	for (int i = startPosition; i < (startPosition + chunkSize); i++) {
 14 | 		if (i < totalPixels) {
 15 | 			int pixelValue = (int)rawPixels[i];
 16 | 			atomicAdd(&histogram[pixelValue], 1);
 17 | 		}
 18 | 	}
 19 | }
 20 | 
 21 | __global__ void kernelComputeClassVariances(double* histogram, double allProbabilitySum, long int totalPixels, double* betweenClassVariance)
 22 | {
 23 | 	int id = blockDim.x * blockIdx.x + threadIdx.x;
 24 | 
 25 | 	double firstClassProbability = 0, secondClassProbability = 0;
 26 | 	double firstClassMean = 0, secondClassMean = 0;
 27 | 	double firstProbabilitySum = 0;
 28 | 
 29 | 	for (int t = 0; t <= id % PngImage::MAX_PIXEL_VALUE; t++) {
 30 | 		firstClassProbability += histogram[t];
 31 | 		firstProbabilitySum += t * histogram[t];
 32 | 	}
 33 | 
 34 | 	secondClassProbability = 1 - firstClassProbability;
 35 | 
 36 | 	firstClassMean = (double)firstProbabilitySum / (double)firstClassProbability;
 37 | 	secondClassMean = (double)(allProbabilitySum - firstProbabilitySum) / (double)secondClassProbability;
 38 | 
 39 | 	betweenClassVariance[id] = firstClassProbability * secondClassProbability * pow((firstClassMean - secondClassMean), 2);
 40 | }
 41 | 
 42 | __global__ void kernelBinarize(unsigned char* rawPixels, long totalPixels, long chunkSize, unsigned char threshold)
 43 | {
 44 | 	int id = blockDim.x * blockIdx.x + threadIdx.x;
 45 | 
 46 | 	int startPosition = id * chunkSize;
 47 | 	for (int i = startPosition; i < (startPosition + chunkSize); i++) {
 48 | 		if (i < totalPixels) {
 49 | 			if ((int)rawPixels[i] >(int)threshold) {
 50 | 				rawPixels[i] = PngImage::COLOR_WHITE;
 51 | 			}
 52 | 			else {
 53 | 				rawPixels[i] = PngImage::COLOR_BLACK;
 54 | 			}
 55 | 		}
 56 | 	}
 57 | }
 58 | 
 59 | CudaOtsuBinarizer::CudaOtsuBinarizer(int threadsPerBlock, int numBlocks, bool drawHistogram, const char* TAG) {
 60 | 	this->threadsPerBlock_ = threadsPerBlock;
 61 | 	this->numBlocks_ = numBlocks;
 62 | 	this->binarizerTimestamp_ = new ExecutionTimestamp();
 63 | 
 64 | 	this->drawHistogram_ = drawHistogram;
 65 | 	this->TAG = TAG;
 66 | }
 67 | 
 68 | CudaOtsuBinarizer::~CudaOtsuBinarizer() {
 69 | 	delete this->binarizerTimestamp_;
 70 | }
 71 | 
 72 | PngImage* CudaOtsuBinarizer::binarize(PngImage * imageToBinarize)
 73 | {
 74 | 	long totalImagePixels = (long)imageToBinarize->getRawPixelData().size();
 75 | 
 76 | 	double* histogram = cudaCalculateHistogram(imageToBinarize->getRawPixelData().data(), totalImagePixels);
 77 | 	cudaDeviceSynchronize();
 78 | 	
 79 | 	if (this->drawHistogram_) {
 80 | 		showHistogram(histogram);
 81 | 	}
 82 | 
 83 | 	unsigned char threshold;
 84 | 	threshold = cudaFindThreshold(histogram, totalImagePixels);
 85 | 	cudaDeviceSynchronize();
 86 | 	printf("\t[%s] Threshold value: %d\n", this->TAG, threshold);
 87 | 
 88 | 	delete histogram;
 89 | 
 90 | 	unsigned char* binarizedRawPixels = cudaBinarize(imageToBinarize->getRawPixelData().data(), totalImagePixels, threshold);
 91 | 	cudaDeviceSynchronize();
 92 | 
 93 | 	std::vector<unsigned char> binarizedVector(&binarizedRawPixels[0], &binarizedRawPixels[totalImagePixels]);
 94 | 
 95 | 	delete binarizedRawPixels;
 96 | 
 97 | 	printf("\n\t[%s] Total calculation time: %.6f seconds \n", this->TAG, binarizerTimestamp_->getExecutionTime());
 98 | 
 99 | 	return new PngImage(
100 | 		imageToBinarize->getFilename(),
101 | 		imageToBinarize->getWidth(),
102 | 		imageToBinarize->getHeight(),
103 | 		binarizedVector
104 | 	);
105 | }
106 | 
107 | std::string CudaOtsuBinarizer::getBinarizerExecutionInfo(std::string fileName)
108 | {
109 | 	return binarizerTimestamp_->toCommaSeparatedRow(fileName, std::string(this->TAG));
110 | }
111 | 
112 | void CudaOtsuBinarizer::showHistogram(double* histogram) {
113 | 	printf("\nHistogram:\n");
114 | 	double value = 0;
115 | 	for (int i = 0; i < PngImage::MAX_PIXEL_VALUE; i++) {
116 | 		value = histogram[i];
117 | 		printf("\tPixel value %d -> %.5f\n", i, value);
118 | 	}
119 | }
120 | 
121 | double* CudaOtsuBinarizer::cudaCalculateHistogram(unsigned char* rawPixels, long totalPixels) {
122 | 	//TODO: check cudaGetDeviceProperties function!
123 | 
124 | 	cudaEvent_t start, stop;
125 | 	cudaEventCreate(&start);
126 | 	cudaEventCreate(&stop);
127 | 
128 | 	unsigned int* hostHistogram = new unsigned int[PngImage::MAX_PIXEL_VALUE];
129 | 	for (int i = 0; i < PngImage::MAX_PIXEL_VALUE; i++) {
130 | 		hostHistogram[i] = 0;
131 | 	}
132 | 
133 | 	unsigned int* deviceHistogram;
134 | 	cudaMalloc((void **)&deviceHistogram, sizeof(unsigned int) * PngImage::MAX_PIXEL_VALUE);
135 | 	cudaMemcpy(deviceHistogram, hostHistogram, sizeof(unsigned int) * PngImage::MAX_PIXEL_VALUE, cudaMemcpyHostToDevice);
136 | 
137 | 	unsigned char* deviceRawPixels;
138 | 	cudaMalloc((void **)&deviceRawPixels, sizeof(unsigned char) * totalPixels);
139 | 	cudaMemcpy(deviceRawPixels, rawPixels, sizeof(unsigned char) * totalPixels, cudaMemcpyHostToDevice);
140 | 
141 | 	long chunkSize = ceil(totalPixels / (threadsPerBlock_ * numBlocks_)) + 1;
142 | 
143 | 	cudaEventRecord(start);
144 | 	kernelCalculateHistogram<<<numBlocks_, threadsPerBlock_>>>(deviceHistogram, deviceRawPixels, chunkSize, totalPixels);
145 | 	cudaEventRecord(stop);
146 | 
147 | 	cudaMemcpy(hostHistogram, deviceHistogram, sizeof(unsigned int) * PngImage::MAX_PIXEL_VALUE, cudaMemcpyDeviceToHost);
148 | 
149 | 	cudaEventSynchronize(stop);
150 | 	float milliseconds = 0;
151 | 	cudaEventElapsedTime(&milliseconds, start, stop);
152 | 	printf("\n\t[%s] Histogram calculated in %.6f milliseconds \n", this->TAG, milliseconds);
153 | 	binarizerTimestamp_->histogramBuildingTimeInSeconds += milliseconds / 1000;
154 | 
155 | 	cudaFree(deviceHistogram);
156 | 	cudaFree(deviceRawPixels);
157 | 
158 | 	double* normalizedHistogram = new double[PngImage::MAX_PIXEL_VALUE];
159 | 	long pixelsSum = 0;
160 | 	for (int v = 0; v < PngImage::MAX_PIXEL_VALUE; v++) {
161 | 		normalizedHistogram[v] = (double)hostHistogram[v] / (double)totalPixels;
162 | 		pixelsSum += hostHistogram[v];
163 | 	}
164 | 	printf("\n\t[%s] Histogram pixels: %d \n", this->TAG, pixelsSum);
165 | 
166 | 	delete hostHistogram;
167 | 
168 | 	return normalizedHistogram;
169 | }
170 | 
171 | unsigned char CudaOtsuBinarizer::cudaFindThreshold(double* histogram, long int totalPixels) {
172 | 	cudaEvent_t start, stop;
173 | 	cudaEventCreate(&start);
174 | 	cudaEventCreate(&stop);
175 | 	int threadsPerBlock = 256;
176 | 	int numBlocks = 1;
177 | 
178 | 	double allProbabilitySum = 0;
179 | 	for (int i = 0; i < PngImage::MAX_PIXEL_VALUE; i++) {
180 | 		allProbabilitySum += i * histogram[i];
181 | 	}
182 | 
183 | 	double* hostBetweenClassVariances = new double[PngImage::MAX_PIXEL_VALUE];
184 | 	for (int i = 0; i < PngImage::MAX_PIXEL_VALUE; i++) {
185 | 		hostBetweenClassVariances[i] = 0;
186 | 	}
187 | 
188 | 	double* deviceHistogram;
189 | 	cudaMalloc((void **)&deviceHistogram, sizeof(double) * PngImage::MAX_PIXEL_VALUE);
190 | 	cudaMemcpy(deviceHistogram, histogram, sizeof(double) * PngImage::MAX_PIXEL_VALUE, cudaMemcpyHostToDevice);
191 | 
192 | 	double* deviceBetweenClassVariances;
193 | 	cudaMalloc((void **)&deviceBetweenClassVariances, sizeof(double) * PngImage::MAX_PIXEL_VALUE);
194 | 	cudaMemcpy(deviceBetweenClassVariances, hostBetweenClassVariances, sizeof(double) * PngImage::MAX_PIXEL_VALUE, cudaMemcpyHostToDevice);
195 | 
196 | 	cudaEventRecord(start);
197 | 	kernelComputeClassVariances<<<numBlocks, threadsPerBlock>>>(deviceHistogram, allProbabilitySum, totalPixels, deviceBetweenClassVariances);
198 | 	cudaEventRecord(stop);
199 | 	cudaMemcpy(hostBetweenClassVariances, deviceBetweenClassVariances, sizeof(double) * PngImage::MAX_PIXEL_VALUE, cudaMemcpyDeviceToHost);
200 | 
201 | 	cudaEventSynchronize(stop);
202 | 	float milliseconds = 0;
203 | 	cudaEventElapsedTime(&milliseconds, start, stop);
204 | 	printf("\n\t[%s] Threshold calculated in %.6f milliseconds \n", this->TAG, milliseconds);
205 | 	binarizerTimestamp_->thresholdFindingTimeInSeconds += milliseconds / 1000;
206 | 
207 | 	cudaFree(deviceHistogram);
208 | 	cudaFree(deviceBetweenClassVariances);
209 | 
210 | 	double maxVariance = 0;
211 | 	unsigned char currentBestThreshold = 0;
212 | 	for (int t = 0; t < PngImage::MAX_PIXEL_VALUE; t++) {
213 | 		if (hostBetweenClassVariances[t] > maxVariance) {
214 | 			currentBestThreshold = (unsigned char)t;
215 | 			maxVariance = hostBetweenClassVariances[t];
216 | 		}
217 | 	}
218 | 
219 | 	delete hostBetweenClassVariances;
220 | 
221 | 	return currentBestThreshold;
222 | }
223 | 
224 | unsigned char* CudaOtsuBinarizer::cudaBinarize(unsigned char * rawPixels, long totalPixels, unsigned char threshold) {
225 | 	cudaEvent_t start, stop;
226 | 	cudaEventCreate(&start);
227 | 	cudaEventCreate(&stop);
228 | 
229 | 	unsigned char* hostRawPixels = new unsigned char[totalPixels];
230 | 
231 | 	unsigned char* deviceRawPixels;
232 | 	cudaMalloc((void **)&deviceRawPixels, sizeof(unsigned char) * totalPixels);
233 | 	cudaMemcpy(deviceRawPixels, rawPixels, totalPixels * sizeof(unsigned char), cudaMemcpyHostToDevice);
234 | 
235 | 	long chunkSize = ceil(totalPixels / (threadsPerBlock_ * numBlocks_)) + 1;
236 | 
237 | 	cudaEventRecord(start);
238 | 	kernelBinarize<<<numBlocks_, threadsPerBlock_>>>(deviceRawPixels, totalPixels, chunkSize, threshold);
239 | 	cudaEventRecord(stop);
240 | 
241 | 	cudaMemcpy(hostRawPixels, deviceRawPixels, sizeof(unsigned char) * totalPixels, cudaMemcpyDeviceToHost);
242 | 
243 | 	cudaEventSynchronize(stop);
244 | 	float milliseconds = 0;
245 | 	cudaEventElapsedTime(&milliseconds, start, stop);
246 | 	printf("\n\t[%s] Binarized in %.6f milliseconds \n", this->TAG, milliseconds);
247 | 	binarizerTimestamp_->binarizationTimeInSeconds += milliseconds / 1000;
248 | 
249 | 	cudaFree(deviceRawPixels);
250 | 
251 | 	return hostRawPixels;
252 | }
253 | 


--------------------------------------------------------------------------------
/src/libs/lodepng.h:
--------------------------------------------------------------------------------
   1 | /*
   2 | LodePNG version 20180326
   3 | 
   4 | Copyright (c) 2005-2018 Lode Vandevenne
   5 | 
   6 | This software is provided 'as-is', without any express or implied
   7 | warranty. In no event will the authors be held liable for any damages
   8 | arising from the use of this software.
   9 | 
  10 | Permission is granted to anyone to use this software for any purpose,
  11 | including commercial applications, and to alter it and redistribute it
  12 | freely, subject to the following restrictions:
  13 | 
  14 |     1. The origin of this software must not be misrepresented; you must not
  15 |     claim that you wrote the original software. If you use this software
  16 |     in a product, an acknowledgment in the product documentation would be
  17 |     appreciated but is not required.
  18 | 
  19 |     2. Altered source versions must be plainly marked as such, and must not be
  20 |     misrepresented as being the original software.
  21 | 
  22 |     3. This notice may not be removed or altered from any source
  23 |     distribution.
  24 | */
  25 | 
  26 | #ifndef LODEPNG_H
  27 | #define LODEPNG_H
  28 | 
  29 | #include <string.h> /*for size_t*/
  30 | 
  31 | extern const char* LODEPNG_VERSION_STRING;
  32 | 
  33 | /*
  34 | The following #defines are used to create code sections. They can be disabled
  35 | to disable code sections, which can give faster compile time and smaller binary.
  36 | The "NO_COMPILE" defines are designed to be used to pass as defines to the
  37 | compiler command to disable them without modifying this header, e.g.
  38 | -DLODEPNG_NO_COMPILE_ZLIB for gcc.
  39 | In addition to those below, you can also define LODEPNG_NO_COMPILE_CRC to
  40 | allow implementing a custom lodepng_crc32.
  41 | */
  42 | /*deflate & zlib. If disabled, you must specify alternative zlib functions in
  43 | the custom_zlib field of the compress and decompress settings*/
  44 | #ifndef LODEPNG_NO_COMPILE_ZLIB
  45 | #define LODEPNG_COMPILE_ZLIB
  46 | #endif
  47 | /*png encoder and png decoder*/
  48 | #ifndef LODEPNG_NO_COMPILE_PNG
  49 | #define LODEPNG_COMPILE_PNG
  50 | #endif
  51 | /*deflate&zlib decoder and png decoder*/
  52 | #ifndef LODEPNG_NO_COMPILE_DECODER
  53 | #define LODEPNG_COMPILE_DECODER
  54 | #endif
  55 | /*deflate&zlib encoder and png encoder*/
  56 | #ifndef LODEPNG_NO_COMPILE_ENCODER
  57 | #define LODEPNG_COMPILE_ENCODER
  58 | #endif
  59 | /*the optional built in harddisk file loading and saving functions*/
  60 | #ifndef LODEPNG_NO_COMPILE_DISK
  61 | #define LODEPNG_COMPILE_DISK
  62 | #endif
  63 | /*support for chunks other than IHDR, IDAT, PLTE, tRNS, IEND: ancillary and unknown chunks*/
  64 | #ifndef LODEPNG_NO_COMPILE_ANCILLARY_CHUNKS
  65 | #define LODEPNG_COMPILE_ANCILLARY_CHUNKS
  66 | #endif
  67 | /*ability to convert error numerical codes to English text string*/
  68 | #ifndef LODEPNG_NO_COMPILE_ERROR_TEXT
  69 | #define LODEPNG_COMPILE_ERROR_TEXT
  70 | #endif
  71 | /*Compile the default allocators (C's free, malloc and realloc). If you disable this,
  72 | you can define the functions lodepng_free, lodepng_malloc and lodepng_realloc in your
  73 | source files with custom allocators.*/
  74 | #ifndef LODEPNG_NO_COMPILE_ALLOCATORS
  75 | #define LODEPNG_COMPILE_ALLOCATORS
  76 | #endif
  77 | /*compile the C++ version (you can disable the C++ wrapper here even when compiling for C++)*/
  78 | #ifdef __cplusplus
  79 | #ifndef LODEPNG_NO_COMPILE_CPP
  80 | #define LODEPNG_COMPILE_CPP
  81 | #endif
  82 | #endif
  83 | 
  84 | #ifdef LODEPNG_COMPILE_CPP
  85 | #include <vector>
  86 | #include <string>
  87 | #endif /*LODEPNG_COMPILE_CPP*/
  88 | 
  89 | #ifdef LODEPNG_COMPILE_PNG
  90 | /*The PNG color types (also used for raw).*/
  91 | typedef enum LodePNGColorType
  92 | {
  93 |   LCT_GREY = 0, /*greyscale: 1,2,4,8,16 bit*/
  94 |   LCT_RGB = 2, /*RGB: 8,16 bit*/
  95 |   LCT_PALETTE = 3, /*palette: 1,2,4,8 bit*/
  96 |   LCT_GREY_ALPHA = 4, /*greyscale with alpha: 8,16 bit*/
  97 |   LCT_RGBA = 6 /*RGB with alpha: 8,16 bit*/
  98 | } LodePNGColorType;
  99 | 
 100 | #ifdef LODEPNG_COMPILE_DECODER
 101 | /*
 102 | Converts PNG data in memory to raw pixel data.
 103 | out: Output parameter. Pointer to buffer that will contain the raw pixel data.
 104 |      After decoding, its size is w * h * (bytes per pixel) bytes larger than
 105 |      initially. Bytes per pixel depends on colortype and bitdepth.
 106 |      Must be freed after usage with free(*out).
 107 |      Note: for 16-bit per channel colors, uses big endian format like PNG does.
 108 | w: Output parameter. Pointer to width of pixel data.
 109 | h: Output parameter. Pointer to height of pixel data.
 110 | in: Memory buffer with the PNG file.
 111 | insize: size of the in buffer.
 112 | colortype: the desired color type for the raw output image. See explanation on PNG color types.
 113 | bitdepth: the desired bit depth for the raw output image. See explanation on PNG color types.
 114 | Return value: LodePNG error code (0 means no error).
 115 | */
 116 | unsigned lodepng_decode_memory(unsigned char** out, unsigned* w, unsigned* h,
 117 |                                const unsigned char* in, size_t insize,
 118 |                                LodePNGColorType colortype, unsigned bitdepth);
 119 | 
 120 | /*Same as lodepng_decode_memory, but always decodes to 32-bit RGBA raw image*/
 121 | unsigned lodepng_decode32(unsigned char** out, unsigned* w, unsigned* h,
 122 |                           const unsigned char* in, size_t insize);
 123 | 
 124 | /*Same as lodepng_decode_memory, but always decodes to 24-bit RGB raw image*/
 125 | unsigned lodepng_decode24(unsigned char** out, unsigned* w, unsigned* h,
 126 |                           const unsigned char* in, size_t insize);
 127 | 
 128 | #ifdef LODEPNG_COMPILE_DISK
 129 | /*
 130 | Load PNG from disk, from file with given name.
 131 | Same as the other decode functions, but instead takes a filename as input.
 132 | */
 133 | unsigned lodepng_decode_file(unsigned char** out, unsigned* w, unsigned* h,
 134 |                              const char* filename,
 135 |                              LodePNGColorType colortype, unsigned bitdepth);
 136 | 
 137 | /*Same as lodepng_decode_file, but always decodes to 32-bit RGBA raw image.*/
 138 | unsigned lodepng_decode32_file(unsigned char** out, unsigned* w, unsigned* h,
 139 |                                const char* filename);
 140 | 
 141 | /*Same as lodepng_decode_file, but always decodes to 24-bit RGB raw image.*/
 142 | unsigned lodepng_decode24_file(unsigned char** out, unsigned* w, unsigned* h,
 143 |                                const char* filename);
 144 | #endif /*LODEPNG_COMPILE_DISK*/
 145 | #endif /*LODEPNG_COMPILE_DECODER*/
 146 | 
 147 | 
 148 | #ifdef LODEPNG_COMPILE_ENCODER
 149 | /*
 150 | Converts raw pixel data into a PNG image in memory. The colortype and bitdepth
 151 |   of the output PNG image cannot be chosen, they are automatically determined
 152 |   by the colortype, bitdepth and content of the input pixel data.
 153 |   Note: for 16-bit per channel colors, needs big endian format like PNG does.
 154 | out: Output parameter. Pointer to buffer that will contain the PNG image data.
 155 |      Must be freed after usage with free(*out).
 156 | outsize: Output parameter. Pointer to the size in bytes of the out buffer.
 157 | image: The raw pixel data to encode. The size of this buffer should be
 158 |        w * h * (bytes per pixel), bytes per pixel depends on colortype and bitdepth.
 159 | w: width of the raw pixel data in pixels.
 160 | h: height of the raw pixel data in pixels.
 161 | colortype: the color type of the raw input image. See explanation on PNG color types.
 162 | bitdepth: the bit depth of the raw input image. See explanation on PNG color types.
 163 | Return value: LodePNG error code (0 means no error).
 164 | */
 165 | unsigned lodepng_encode_memory(unsigned char** out, size_t* outsize,
 166 |                                const unsigned char* image, unsigned w, unsigned h,
 167 |                                LodePNGColorType colortype, unsigned bitdepth);
 168 | 
 169 | /*Same as lodepng_encode_memory, but always encodes from 32-bit RGBA raw image.*/
 170 | unsigned lodepng_encode32(unsigned char** out, size_t* outsize,
 171 |                           const unsigned char* image, unsigned w, unsigned h);
 172 | 
 173 | /*Same as lodepng_encode_memory, but always encodes from 24-bit RGB raw image.*/
 174 | unsigned lodepng_encode24(unsigned char** out, size_t* outsize,
 175 |                           const unsigned char* image, unsigned w, unsigned h);
 176 | 
 177 | #ifdef LODEPNG_COMPILE_DISK
 178 | /*
 179 | Converts raw pixel data into a PNG file on disk.
 180 | Same as the other encode functions, but instead takes a filename as output.
 181 | NOTE: This overwrites existing files without warning!
 182 | */
 183 | unsigned lodepng_encode_file(const char* filename,
 184 |                              const unsigned char* image, unsigned w, unsigned h,
 185 |                              LodePNGColorType colortype, unsigned bitdepth);
 186 | 
 187 | /*Same as lodepng_encode_file, but always encodes from 32-bit RGBA raw image.*/
 188 | unsigned lodepng_encode32_file(const char* filename,
 189 |                                const unsigned char* image, unsigned w, unsigned h);
 190 | 
 191 | /*Same as lodepng_encode_file, but always encodes from 24-bit RGB raw image.*/
 192 | unsigned lodepng_encode24_file(const char* filename,
 193 |                                const unsigned char* image, unsigned w, unsigned h);
 194 | #endif /*LODEPNG_COMPILE_DISK*/
 195 | #endif /*LODEPNG_COMPILE_ENCODER*/
 196 | 
 197 | 
 198 | #ifdef LODEPNG_COMPILE_CPP
 199 | namespace lodepng
 200 | {
 201 | #ifdef LODEPNG_COMPILE_DECODER
 202 | /*Same as lodepng_decode_memory, but decodes to an std::vector. The colortype
 203 | is the format to output the pixels to. Default is RGBA 8-bit per channel.*/
 204 | unsigned decode(std::vector<unsigned char>& out, unsigned& w, unsigned& h,
 205 |                 const unsigned char* in, size_t insize,
 206 |                 LodePNGColorType colortype = LCT_RGBA, unsigned bitdepth = 8);
 207 | unsigned decode(std::vector<unsigned char>& out, unsigned& w, unsigned& h,
 208 |                 const std::vector<unsigned char>& in,
 209 |                 LodePNGColorType colortype = LCT_RGBA, unsigned bitdepth = 8);
 210 | #ifdef LODEPNG_COMPILE_DISK
 211 | /*
 212 | Converts PNG file from disk to raw pixel data in memory.
 213 | Same as the other decode functions, but instead takes a filename as input.
 214 | */
 215 | unsigned decode(std::vector<unsigned char>& out, unsigned& w, unsigned& h,
 216 |                 const std::string& filename,
 217 |                 LodePNGColorType colortype = LCT_RGBA, unsigned bitdepth = 8);
 218 | #endif /* LODEPNG_COMPILE_DISK */
 219 | #endif /* LODEPNG_COMPILE_DECODER */
 220 | 
 221 | #ifdef LODEPNG_COMPILE_ENCODER
 222 | /*Same as lodepng_encode_memory, but encodes to an std::vector. colortype
 223 | is that of the raw input data. The output PNG color type will be auto chosen.*/
 224 | unsigned encode(std::vector<unsigned char>& out,
 225 |                 const unsigned char* in, unsigned w, unsigned h,
 226 |                 LodePNGColorType colortype = LCT_RGBA, unsigned bitdepth = 8);
 227 | unsigned encode(std::vector<unsigned char>& out,
 228 |                 const std::vector<unsigned char>& in, unsigned w, unsigned h,
 229 |                 LodePNGColorType colortype = LCT_RGBA, unsigned bitdepth = 8);
 230 | #ifdef LODEPNG_COMPILE_DISK
 231 | /*
 232 | Converts 32-bit RGBA raw pixel data into a PNG file on disk.
 233 | Same as the other encode functions, but instead takes a filename as output.
 234 | NOTE: This overwrites existing files without warning!
 235 | */
 236 | unsigned encode(const std::string& filename,
 237 |                 const unsigned char* in, unsigned w, unsigned h,
 238 |                 LodePNGColorType colortype = LCT_RGBA, unsigned bitdepth = 8);
 239 | unsigned encode(const std::string& filename,
 240 |                 const std::vector<unsigned char>& in, unsigned w, unsigned h,
 241 |                 LodePNGColorType colortype = LCT_RGBA, unsigned bitdepth = 8);
 242 | #endif /* LODEPNG_COMPILE_DISK */
 243 | #endif /* LODEPNG_COMPILE_ENCODER */
 244 | } /* namespace lodepng */
 245 | #endif /*LODEPNG_COMPILE_CPP*/
 246 | #endif /*LODEPNG_COMPILE_PNG*/
 247 | 
 248 | #ifdef LODEPNG_COMPILE_ERROR_TEXT
 249 | /*Returns an English description of the numerical error code.*/
 250 | const char* lodepng_error_text(unsigned code);
 251 | #endif /*LODEPNG_COMPILE_ERROR_TEXT*/
 252 | 
 253 | #ifdef LODEPNG_COMPILE_DECODER
 254 | /*Settings for zlib decompression*/
 255 | typedef struct LodePNGDecompressSettings LodePNGDecompressSettings;
 256 | struct LodePNGDecompressSettings
 257 | {
 258 |   /* Check LodePNGDecoderSettings for more ignorable errors */
 259 |   unsigned ignore_adler32; /*if 1, continue and don't give an error message if the Adler32 checksum is corrupted*/
 260 | 
 261 |   /*use custom zlib decoder instead of built in one (default: null)*/
 262 |   unsigned (*custom_zlib)(unsigned char**, size_t*,
 263 |                           const unsigned char*, size_t,
 264 |                           const LodePNGDecompressSettings*);
 265 |   /*use custom deflate decoder instead of built in one (default: null)
 266 |   if custom_zlib is used, custom_deflate is ignored since only the built in
 267 |   zlib function will call custom_deflate*/
 268 |   unsigned (*custom_inflate)(unsigned char**, size_t*,
 269 |                              const unsigned char*, size_t,
 270 |                              const LodePNGDecompressSettings*);
 271 | 
 272 |   const void* custom_context; /*optional custom settings for custom functions*/
 273 | };
 274 | 
 275 | extern const LodePNGDecompressSettings lodepng_default_decompress_settings;
 276 | void lodepng_decompress_settings_init(LodePNGDecompressSettings* settings);
 277 | #endif /*LODEPNG_COMPILE_DECODER*/
 278 | 
 279 | #ifdef LODEPNG_COMPILE_ENCODER
 280 | /*
 281 | Settings for zlib compression. Tweaking these settings tweaks the balance
 282 | between speed and compression ratio.
 283 | */
 284 | typedef struct LodePNGCompressSettings LodePNGCompressSettings;
 285 | struct LodePNGCompressSettings /*deflate = compress*/
 286 | {
 287 |   /*LZ77 related settings*/
 288 |   unsigned btype; /*the block type for LZ (0, 1, 2 or 3, see zlib standard). Should be 2 for proper compression.*/
 289 |   unsigned use_lz77; /*whether or not to use LZ77. Should be 1 for proper compression.*/
 290 |   unsigned windowsize; /*must be a power of two <= 32768. higher compresses more but is slower. Default value: 2048.*/
 291 |   unsigned minmatch; /*mininum lz77 length. 3 is normally best, 6 can be better for some PNGs. Default: 0*/
 292 |   unsigned nicematch; /*stop searching if >= this length found. Set to 258 for best compression. Default: 128*/
 293 |   unsigned lazymatching; /*use lazy matching: better compression but a bit slower. Default: true*/
 294 | 
 295 |   /*use custom zlib encoder instead of built in one (default: null)*/
 296 |   unsigned (*custom_zlib)(unsigned char**, size_t*,
 297 |                           const unsigned char*, size_t,
 298 |                           const LodePNGCompressSettings*);
 299 |   /*use custom deflate encoder instead of built in one (default: null)
 300 |   if custom_zlib is used, custom_deflate is ignored since only the built in
 301 |   zlib function will call custom_deflate*/
 302 |   unsigned (*custom_deflate)(unsigned char**, size_t*,
 303 |                              const unsigned char*, size_t,
 304 |                              const LodePNGCompressSettings*);
 305 | 
 306 |   const void* custom_context; /*optional custom settings for custom functions*/
 307 | };
 308 | 
 309 | extern const LodePNGCompressSettings lodepng_default_compress_settings;
 310 | void lodepng_compress_settings_init(LodePNGCompressSettings* settings);
 311 | #endif /*LODEPNG_COMPILE_ENCODER*/
 312 | 
 313 | #ifdef LODEPNG_COMPILE_PNG
 314 | /*
 315 | Color mode of an image. Contains all information required to decode the pixel
 316 | bits to RGBA colors. This information is the same as used in the PNG file
 317 | format, and is used both for PNG and raw image data in LodePNG.
 318 | */
 319 | typedef struct LodePNGColorMode
 320 | {
 321 |   /*header (IHDR)*/
 322 |   LodePNGColorType colortype; /*color type, see PNG standard or documentation further in this header file*/
 323 |   unsigned bitdepth;  /*bits per sample, see PNG standard or documentation further in this header file*/
 324 | 
 325 |   /*
 326 |   palette (PLTE and tRNS)
 327 | 
 328 |   Dynamically allocated with the colors of the palette, including alpha.
 329 |   When encoding a PNG, to store your colors in the palette of the LodePNGColorMode, first use
 330 |   lodepng_palette_clear, then for each color use lodepng_palette_add.
 331 |   If you encode an image without alpha with palette, don't forget to put value 255 in each A byte of the palette.
 332 | 
 333 |   When decoding, by default you can ignore this palette, since LodePNG already
 334 |   fills the palette colors in the pixels of the raw RGBA output.
 335 | 
 336 |   The palette is only supported for color type 3.
 337 |   */
 338 |   unsigned char* palette; /*palette in RGBARGBA... order. When allocated, must be either 0, or have size 1024*/
 339 |   size_t palettesize; /*palette size in number of colors (amount of bytes is 4 * palettesize)*/
 340 | 
 341 |   /*
 342 |   transparent color key (tRNS)
 343 | 
 344 |   This color uses the same bit depth as the bitdepth value in this struct, which can be 1-bit to 16-bit.
 345 |   For greyscale PNGs, r, g and b will all 3 be set to the same.
 346 | 
 347 |   When decoding, by default you can ignore this information, since LodePNG sets
 348 |   pixels with this key to transparent already in the raw RGBA output.
 349 | 
 350 |   The color key is only supported for color types 0 and 2.
 351 |   */
 352 |   unsigned key_defined; /*is a transparent color key given? 0 = false, 1 = true*/
 353 |   unsigned key_r;       /*red/greyscale component of color key*/
 354 |   unsigned key_g;       /*green component of color key*/
 355 |   unsigned key_b;       /*blue component of color key*/
 356 | } LodePNGColorMode;
 357 | 
 358 | /*init, cleanup and copy functions to use with this struct*/
 359 | void lodepng_color_mode_init(LodePNGColorMode* info);
 360 | void lodepng_color_mode_cleanup(LodePNGColorMode* info);
 361 | /*return value is error code (0 means no error)*/
 362 | unsigned lodepng_color_mode_copy(LodePNGColorMode* dest, const LodePNGColorMode* source);
 363 | 
 364 | void lodepng_palette_clear(LodePNGColorMode* info);
 365 | /*add 1 color to the palette*/
 366 | unsigned lodepng_palette_add(LodePNGColorMode* info,
 367 |                              unsigned char r, unsigned char g, unsigned char b, unsigned char a);
 368 | 
 369 | /*get the total amount of bits per pixel, based on colortype and bitdepth in the struct*/
 370 | unsigned lodepng_get_bpp(const LodePNGColorMode* info);
 371 | /*get the amount of color channels used, based on colortype in the struct.
 372 | If a palette is used, it counts as 1 channel.*/
 373 | unsigned lodepng_get_channels(const LodePNGColorMode* info);
 374 | /*is it a greyscale type? (only colortype 0 or 4)*/
 375 | unsigned lodepng_is_greyscale_type(const LodePNGColorMode* info);
 376 | /*has it got an alpha channel? (only colortype 2 or 6)*/
 377 | unsigned lodepng_is_alpha_type(const LodePNGColorMode* info);
 378 | /*has it got a palette? (only colortype 3)*/
 379 | unsigned lodepng_is_palette_type(const LodePNGColorMode* info);
 380 | /*only returns true if there is a palette and there is a value in the palette with alpha < 255.
 381 | Loops through the palette to check this.*/
 382 | unsigned lodepng_has_palette_alpha(const LodePNGColorMode* info);
 383 | /*
 384 | Check if the given color info indicates the possibility of having non-opaque pixels in the PNG image.
 385 | Returns true if the image can have translucent or invisible pixels (it still be opaque if it doesn't use such pixels).
 386 | Returns false if the image can only have opaque pixels.
 387 | In detail, it returns true only if it's a color type with alpha, or has a palette with non-opaque values,
 388 | or if "key_defined" is true.
 389 | */
 390 | unsigned lodepng_can_have_alpha(const LodePNGColorMode* info);
 391 | /*Returns the byte size of a raw image buffer with given width, height and color mode*/
 392 | size_t lodepng_get_raw_size(unsigned w, unsigned h, const LodePNGColorMode* color);
 393 | 
 394 | #ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
 395 | /*The information of a Time chunk in PNG.*/
 396 | typedef struct LodePNGTime
 397 | {
 398 |   unsigned year;    /*2 bytes used (0-65535)*/
 399 |   unsigned month;   /*1-12*/
 400 |   unsigned day;     /*1-31*/
 401 |   unsigned hour;    /*0-23*/
 402 |   unsigned minute;  /*0-59*/
 403 |   unsigned second;  /*0-60 (to allow for leap seconds)*/
 404 | } LodePNGTime;
 405 | #endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
 406 | 
 407 | /*Information about the PNG image, except pixels, width and height.*/
 408 | typedef struct LodePNGInfo
 409 | {
 410 |   /*header (IHDR), palette (PLTE) and transparency (tRNS) chunks*/
 411 |   unsigned compression_method;/*compression method of the original file. Always 0.*/
 412 |   unsigned filter_method;     /*filter method of the original file*/
 413 |   unsigned interlace_method;  /*interlace method of the original file*/
 414 |   LodePNGColorMode color;     /*color type and bits, palette and transparency of the PNG file*/
 415 | 
 416 | #ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
 417 |   /*
 418 |   suggested background color chunk (bKGD)
 419 |   This color uses the same color mode as the PNG (except alpha channel), which can be 1-bit to 16-bit.
 420 | 
 421 |   For greyscale PNGs, r, g and b will all 3 be set to the same. When encoding
 422 |   the encoder writes the red one. For palette PNGs: When decoding, the RGB value
 423 |   will be stored, not a palette index. But when encoding, specify the index of
 424 |   the palette in background_r, the other two are then ignored.
 425 | 
 426 |   The decoder does not use this background color to edit the color of pixels.
 427 |   */
 428 |   unsigned background_defined; /*is a suggested background color given?*/
 429 |   unsigned background_r;       /*red component of suggested background color*/
 430 |   unsigned background_g;       /*green component of suggested background color*/
 431 |   unsigned background_b;       /*blue component of suggested background color*/
 432 | 
 433 |   /*
 434 |   non-international text chunks (tEXt and zTXt)
 435 | 
 436 |   The char** arrays each contain num strings. The actual messages are in
 437 |   text_strings, while text_keys are keywords that give a short description what
 438 |   the actual text represents, e.g. Title, Author, Description, or anything else.
 439 | 
 440 |   A keyword is minimum 1 character and maximum 79 characters long. It's
 441 |   discouraged to use a single line length longer than 79 characters for texts.
 442 | 
 443 |   Don't allocate these text buffers yourself. Use the init/cleanup functions
 444 |   correctly and use lodepng_add_text and lodepng_clear_text.
 445 |   */
 446 |   size_t text_num; /*the amount of texts in these char** buffers (there may be more texts in itext)*/
 447 |   char** text_keys; /*the keyword of a text chunk (e.g. "Comment")*/
 448 |   char** text_strings; /*the actual text*/
 449 | 
 450 |   /*
 451 |   international text chunks (iTXt)
 452 |   Similar to the non-international text chunks, but with additional strings
 453 |   "langtags" and "transkeys".
 454 |   */
 455 |   size_t itext_num; /*the amount of international texts in this PNG*/
 456 |   char** itext_keys; /*the English keyword of the text chunk (e.g. "Comment")*/
 457 |   char** itext_langtags; /*language tag for this text's language, ISO/IEC 646 string, e.g. ISO 639 language tag*/
 458 |   char** itext_transkeys; /*keyword translated to the international language - UTF-8 string*/
 459 |   char** itext_strings; /*the actual international text - UTF-8 string*/
 460 | 
 461 |   /*time chunk (tIME)*/
 462 |   unsigned time_defined; /*set to 1 to make the encoder generate a tIME chunk*/
 463 |   LodePNGTime time;
 464 | 
 465 |   /*phys chunk (pHYs)*/
 466 |   unsigned phys_defined; /*if 0, there is no pHYs chunk and the values below are undefined, if 1 else there is one*/
 467 |   unsigned phys_x; /*pixels per unit in x direction*/
 468 |   unsigned phys_y; /*pixels per unit in y direction*/
 469 |   unsigned phys_unit; /*may be 0 (unknown unit) or 1 (metre)*/
 470 | 
 471 |   /*
 472 |   unknown chunks
 473 |   There are 3 buffers, one for each position in the PNG where unknown chunks can appear
 474 |   each buffer contains all unknown chunks for that position consecutively
 475 |   The 3 buffers are the unknown chunks between certain critical chunks:
 476 |   0: IHDR-PLTE, 1: PLTE-IDAT, 2: IDAT-IEND
 477 |   Do not allocate or traverse this data yourself. Use the chunk traversing functions declared
 478 |   later, such as lodepng_chunk_next and lodepng_chunk_append, to read/write this struct.
 479 |   */
 480 |   unsigned char* unknown_chunks_data[3];
 481 |   size_t unknown_chunks_size[3]; /*size in bytes of the unknown chunks, given for protection*/
 482 | #endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
 483 | } LodePNGInfo;
 484 | 
 485 | /*init, cleanup and copy functions to use with this struct*/
 486 | void lodepng_info_init(LodePNGInfo* info);
 487 | void lodepng_info_cleanup(LodePNGInfo* info);
 488 | /*return value is error code (0 means no error)*/
 489 | unsigned lodepng_info_copy(LodePNGInfo* dest, const LodePNGInfo* source);
 490 | 
 491 | #ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
 492 | void lodepng_clear_text(LodePNGInfo* info); /*use this to clear the texts again after you filled them in*/
 493 | unsigned lodepng_add_text(LodePNGInfo* info, const char* key, const char* str); /*push back both texts at once*/
 494 | 
 495 | void lodepng_clear_itext(LodePNGInfo* info); /*use this to clear the itexts again after you filled them in*/
 496 | unsigned lodepng_add_itext(LodePNGInfo* info, const char* key, const char* langtag,
 497 |                            const char* transkey, const char* str); /*push back the 4 texts of 1 chunk at once*/
 498 | #endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
 499 | 
 500 | /*
 501 | Converts raw buffer from one color type to another color type, based on
 502 | LodePNGColorMode structs to describe the input and output color type.
 503 | See the reference manual at the end of this header file to see which color conversions are supported.
 504 | return value = LodePNG error code (0 if all went ok, an error if the conversion isn't supported)
 505 | The out buffer must have size (w * h * bpp + 7) / 8, where bpp is the bits per pixel
 506 | of the output color type (lodepng_get_bpp).
 507 | For < 8 bpp images, there should not be padding bits at the end of scanlines.
 508 | For 16-bit per channel colors, uses big endian format like PNG does.
 509 | Return value is LodePNG error code
 510 | */
 511 | unsigned lodepng_convert(unsigned char* out, const unsigned char* in,
 512 |                          const LodePNGColorMode* mode_out, const LodePNGColorMode* mode_in,
 513 |                          unsigned w, unsigned h);
 514 | 
 515 | #ifdef LODEPNG_COMPILE_DECODER
 516 | /*
 517 | Settings for the decoder. This contains settings for the PNG and the Zlib
 518 | decoder, but not the Info settings from the Info structs.
 519 | */
 520 | typedef struct LodePNGDecoderSettings
 521 | {
 522 |   LodePNGDecompressSettings zlibsettings; /*in here is the setting to ignore Adler32 checksums*/
 523 | 
 524 |   /* Check LodePNGDecompressSettings for more ignorable errors */
 525 |   unsigned ignore_crc; /*ignore CRC checksums*/
 526 |   unsigned ignore_critical; /*ignore unknown critical chunks*/
 527 |   unsigned ignore_end; /*ignore issues at end of file if possible (missing IEND chunk, too large chunk, ...)*/
 528 | 
 529 |   unsigned color_convert; /*whether to convert the PNG to the color type you want. Default: yes*/
 530 | 
 531 | #ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
 532 |   unsigned read_text_chunks; /*if false but remember_unknown_chunks is true, they're stored in the unknown chunks*/
 533 |   /*store all bytes from unknown chunks in the LodePNGInfo (off by default, useful for a png editor)*/
 534 |   unsigned remember_unknown_chunks;
 535 | #endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
 536 | } LodePNGDecoderSettings;
 537 | 
 538 | void lodepng_decoder_settings_init(LodePNGDecoderSettings* settings);
 539 | #endif /*LODEPNG_COMPILE_DECODER*/
 540 | 
 541 | #ifdef LODEPNG_COMPILE_ENCODER
 542 | /*automatically use color type with less bits per pixel if losslessly possible. Default: AUTO*/
 543 | typedef enum LodePNGFilterStrategy
 544 | {
 545 |   /*every filter at zero*/
 546 |   LFS_ZERO,
 547 |   /*Use filter that gives minimum sum, as described in the official PNG filter heuristic.*/
 548 |   LFS_MINSUM,
 549 |   /*Use the filter type that gives smallest Shannon entropy for this scanline. Depending
 550 |   on the image, this is better or worse than minsum.*/
 551 |   LFS_ENTROPY,
 552 |   /*
 553 |   Brute-force-search PNG filters by compressing each filter for each scanline.
 554 |   Experimental, very slow, and only rarely gives better compression than MINSUM.
 555 |   */
 556 |   LFS_BRUTE_FORCE,
 557 |   /*use predefined_filters buffer: you specify the filter type for each scanline*/
 558 |   LFS_PREDEFINED
 559 | } LodePNGFilterStrategy;
 560 | 
 561 | /*Gives characteristics about the colors of the image, which helps decide which color model to use for encoding.
 562 | Used internally by default if "auto_convert" is enabled. Public because it's useful for custom algorithms.*/
 563 | typedef struct LodePNGColorProfile
 564 | {
 565 |   unsigned colored; /*not greyscale*/
 566 |   unsigned key; /*image is not opaque and color key is possible instead of full alpha*/
 567 |   unsigned short key_r; /*key values, always as 16-bit, in 8-bit case the byte is duplicated, e.g. 65535 means 255*/
 568 |   unsigned short key_g;
 569 |   unsigned short key_b;
 570 |   unsigned alpha; /*image is not opaque and alpha channel or alpha palette required*/
 571 |   unsigned numcolors; /*amount of colors, up to 257. Not valid if bits == 16.*/
 572 |   unsigned char palette[1024]; /*Remembers up to the first 256 RGBA colors, in no particular order*/
 573 |   unsigned bits; /*bits per channel (not for palette). 1,2 or 4 for greyscale only. 16 if 16-bit per channel required.*/
 574 | } LodePNGColorProfile;
 575 | 
 576 | void lodepng_color_profile_init(LodePNGColorProfile* profile);
 577 | 
 578 | /*Get a LodePNGColorProfile of the image.*/
 579 | unsigned lodepng_get_color_profile(LodePNGColorProfile* profile,
 580 |                                    const unsigned char* image, unsigned w, unsigned h,
 581 |                                    const LodePNGColorMode* mode_in);
 582 | /*The function LodePNG uses internally to decide the PNG color with auto_convert.
 583 | Chooses an optimal color model, e.g. grey if only grey pixels, palette if < 256 colors, ...*/
 584 | unsigned lodepng_auto_choose_color(LodePNGColorMode* mode_out,
 585 |                                    const unsigned char* image, unsigned w, unsigned h,
 586 |                                    const LodePNGColorMode* mode_in);
 587 | 
 588 | /*Settings for the encoder.*/
 589 | typedef struct LodePNGEncoderSettings
 590 | {
 591 |   LodePNGCompressSettings zlibsettings; /*settings for the zlib encoder, such as window size, ...*/
 592 | 
 593 |   unsigned auto_convert; /*automatically choose output PNG color type. Default: true*/
 594 | 
 595 |   /*If true, follows the official PNG heuristic: if the PNG uses a palette or lower than
 596 |   8 bit depth, set all filters to zero. Otherwise use the filter_strategy. Note that to
 597 |   completely follow the official PNG heuristic, filter_palette_zero must be true and
 598 |   filter_strategy must be LFS_MINSUM*/
 599 |   unsigned filter_palette_zero;
 600 |   /*Which filter strategy to use when not using zeroes due to filter_palette_zero.
 601 |   Set filter_palette_zero to 0 to ensure always using your chosen strategy. Default: LFS_MINSUM*/
 602 |   LodePNGFilterStrategy filter_strategy;
 603 |   /*used if filter_strategy is LFS_PREDEFINED. In that case, this must point to a buffer with
 604 |   the same length as the amount of scanlines in the image, and each value must <= 5. You
 605 |   have to cleanup this buffer, LodePNG will never free it. Don't forget that filter_palette_zero
 606 |   must be set to 0 to ensure this is also used on palette or low bitdepth images.*/
 607 |   const unsigned char* predefined_filters;
 608 | 
 609 |   /*force creating a PLTE chunk if colortype is 2 or 6 (= a suggested palette).
 610 |   If colortype is 3, PLTE is _always_ created.*/
 611 |   unsigned force_palette;
 612 | #ifdef LODEPNG_COMPILE_ANCILLARY_CHUNKS
 613 |   /*add LodePNG identifier and version as a text chunk, for debugging*/
 614 |   unsigned add_id;
 615 |   /*encode text chunks as zTXt chunks instead of tEXt chunks, and use compression in iTXt chunks*/
 616 |   unsigned text_compression;
 617 | #endif /*LODEPNG_COMPILE_ANCILLARY_CHUNKS*/
 618 | } LodePNGEncoderSettings;
 619 | 
 620 | void lodepng_encoder_settings_init(LodePNGEncoderSettings* settings);
 621 | #endif /*LODEPNG_COMPILE_ENCODER*/
 622 | 
 623 | 
 624 | #if defined(LODEPNG_COMPILE_DECODER) || defined(LODEPNG_COMPILE_ENCODER)
 625 | /*The settings, state and information for extended encoding and decoding.*/
 626 | typedef struct LodePNGState
 627 | {
 628 | #ifdef LODEPNG_COMPILE_DECODER
 629 |   LodePNGDecoderSettings decoder; /*the decoding settings*/
 630 | #endif /*LODEPNG_COMPILE_DECODER*/
 631 | #ifdef LODEPNG_COMPILE_ENCODER
 632 |   LodePNGEncoderSettings encoder; /*the encoding settings*/
 633 | #endif /*LODEPNG_COMPILE_ENCODER*/
 634 |   LodePNGColorMode info_raw; /*specifies the format in which you would like to get the raw pixel buffer*/
 635 |   LodePNGInfo info_png; /*info of the PNG image obtained after decoding*/
 636 |   unsigned error;
 637 | #ifdef LODEPNG_COMPILE_CPP
 638 |   /* For the lodepng::State subclass. */
 639 |   virtual ~LodePNGState(){}
 640 | #endif
 641 | } LodePNGState;
 642 | 
 643 | /*init, cleanup and copy functions to use with this struct*/
 644 | void lodepng_state_init(LodePNGState* state);
 645 | void lodepng_state_cleanup(LodePNGState* state);
 646 | void lodepng_state_copy(LodePNGState* dest, const LodePNGState* source);
 647 | #endif /* defined(LODEPNG_COMPILE_DECODER) || defined(LODEPNG_COMPILE_ENCODER) */
 648 | 
 649 | #ifdef LODEPNG_COMPILE_DECODER
 650 | /*
 651 | Same as lodepng_decode_memory, but uses a LodePNGState to allow custom settings and
 652 | getting much more information about the PNG image and color mode.
 653 | */
 654 | unsigned lodepng_decode(unsigned char** out, unsigned* w, unsigned* h,
 655 |                         LodePNGState* state,
 656 |                         const unsigned char* in, size_t insize);
 657 | 
 658 | /*
 659 | Read the PNG header, but not the actual data. This returns only the information
 660 | that is in the header chunk of the PNG, such as width, height and color type. The
 661 | information is placed in the info_png field of the LodePNGState.
 662 | */
 663 | unsigned lodepng_inspect(unsigned* w, unsigned* h,
 664 |                          LodePNGState* state,
 665 |                          const unsigned char* in, size_t insize);
 666 | #endif /*LODEPNG_COMPILE_DECODER*/
 667 | 
 668 | 
 669 | #ifdef LODEPNG_COMPILE_ENCODER
 670 | /*This function allocates the out buffer with standard malloc and stores the size in *outsize.*/
 671 | unsigned lodepng_encode(unsigned char** out, size_t* outsize,
 672 |                         const unsigned char* image, unsigned w, unsigned h,
 673 |                         LodePNGState* state);
 674 | #endif /*LODEPNG_COMPILE_ENCODER*/
 675 | 
 676 | /*
 677 | The lodepng_chunk functions are normally not needed, except to traverse the
 678 | unknown chunks stored in the LodePNGInfo struct, or add new ones to it.
 679 | It also allows traversing the chunks of an encoded PNG file yourself.
 680 | 
 681 | PNG standard chunk naming conventions:
 682 | First byte: uppercase = critical, lowercase = ancillary
 683 | Second byte: uppercase = public, lowercase = private
 684 | Third byte: must be uppercase
 685 | Fourth byte: uppercase = unsafe to copy, lowercase = safe to copy
 686 | */
 687 | 
 688 | /*
 689 | Gets the length of the data of the chunk. Total chunk length has 12 bytes more.
 690 | There must be at least 4 bytes to read from. If the result value is too large,
 691 | it may be corrupt data.
 692 | */
 693 | unsigned lodepng_chunk_length(const unsigned char* chunk);
 694 | 
 695 | /*puts the 4-byte type in null terminated string*/
 696 | void lodepng_chunk_type(char type[5], const unsigned char* chunk);
 697 | 
 698 | /*check if the type is the given type*/
 699 | unsigned char lodepng_chunk_type_equals(const unsigned char* chunk, const char* type);
 700 | 
 701 | /*0: it's one of the critical chunk types, 1: it's an ancillary chunk (see PNG standard)*/
 702 | unsigned char lodepng_chunk_ancillary(const unsigned char* chunk);
 703 | 
 704 | /*0: public, 1: private (see PNG standard)*/
 705 | unsigned char lodepng_chunk_private(const unsigned char* chunk);
 706 | 
 707 | /*0: the chunk is unsafe to copy, 1: the chunk is safe to copy (see PNG standard)*/
 708 | unsigned char lodepng_chunk_safetocopy(const unsigned char* chunk);
 709 | 
 710 | /*get pointer to the data of the chunk, where the input points to the header of the chunk*/
 711 | unsigned char* lodepng_chunk_data(unsigned char* chunk);
 712 | const unsigned char* lodepng_chunk_data_const(const unsigned char* chunk);
 713 | 
 714 | /*returns 0 if the crc is correct, 1 if it's incorrect (0 for OK as usual!)*/
 715 | unsigned lodepng_chunk_check_crc(const unsigned char* chunk);
 716 | 
 717 | /*generates the correct CRC from the data and puts it in the last 4 bytes of the chunk*/
 718 | void lodepng_chunk_generate_crc(unsigned char* chunk);
 719 | 
 720 | /*iterate to next chunks. don't use on IEND chunk, as there is no next chunk then*/
 721 | unsigned char* lodepng_chunk_next(unsigned char* chunk);
 722 | const unsigned char* lodepng_chunk_next_const(const unsigned char* chunk);
 723 | 
 724 | /*
 725 | Appends chunk to the data in out. The given chunk should already have its chunk header.
 726 | The out variable and outlength are updated to reflect the new reallocated buffer.
 727 | Returns error code (0 if it went ok)
 728 | */
 729 | unsigned lodepng_chunk_append(unsigned char** out, size_t* outlength, const unsigned char* chunk);
 730 | 
 731 | /*
 732 | Appends new chunk to out. The chunk to append is given by giving its length, type
 733 | and data separately. The type is a 4-letter string.
 734 | The out variable and outlength are updated to reflect the new reallocated buffer.
 735 | Returne error code (0 if it went ok)
 736 | */
 737 | unsigned lodepng_chunk_create(unsigned char** out, size_t* outlength, unsigned length,
 738 |                               const char* type, const unsigned char* data);
 739 | 
 740 | 
 741 | /*Calculate CRC32 of buffer*/
 742 | unsigned lodepng_crc32(const unsigned char* buf, size_t len);
 743 | #endif /*LODEPNG_COMPILE_PNG*/
 744 | 
 745 | 
 746 | #ifdef LODEPNG_COMPILE_ZLIB
 747 | /*
 748 | This zlib part can be used independently to zlib compress and decompress a
 749 | buffer. It cannot be used to create gzip files however, and it only supports the
 750 | part of zlib that is required for PNG, it does not support dictionaries.
 751 | */
 752 | 
 753 | #ifdef LODEPNG_COMPILE_DECODER
 754 | /*Inflate a buffer. Inflate is the decompression step of deflate. Out buffer must be freed after use.*/
 755 | unsigned lodepng_inflate(unsigned char** out, size_t* outsize,
 756 |                          const unsigned char* in, size_t insize,
 757 |                          const LodePNGDecompressSettings* settings);
 758 | 
 759 | /*
 760 | Decompresses Zlib data. Reallocates the out buffer and appends the data. The
 761 | data must be according to the zlib specification.
 762 | Either, *out must be NULL and *outsize must be 0, or, *out must be a valid
 763 | buffer and *outsize its size in bytes. out must be freed by user after usage.
 764 | */
 765 | unsigned lodepng_zlib_decompress(unsigned char** out, size_t* outsize,
 766 |                                  const unsigned char* in, size_t insize,
 767 |                                  const LodePNGDecompressSettings* settings);
 768 | #endif /*LODEPNG_COMPILE_DECODER*/
 769 | 
 770 | #ifdef LODEPNG_COMPILE_ENCODER
 771 | /*
 772 | Compresses data with Zlib. Reallocates the out buffer and appends the data.
 773 | Zlib adds a small header and trailer around the deflate data.
 774 | The data is output in the format of the zlib specification.
 775 | Either, *out must be NULL and *outsize must be 0, or, *out must be a valid
 776 | buffer and *outsize its size in bytes. out must be freed by user after usage.
 777 | */
 778 | unsigned lodepng_zlib_compress(unsigned char** out, size_t* outsize,
 779 |                                const unsigned char* in, size_t insize,
 780 |                                const LodePNGCompressSettings* settings);
 781 | 
 782 | /*
 783 | Find length-limited Huffman code for given frequencies. This function is in the
 784 | public interface only for tests, it's used internally by lodepng_deflate.
 785 | */
 786 | unsigned lodepng_huffman_code_lengths(unsigned* lengths, const unsigned* frequencies,
 787 |                                       size_t numcodes, unsigned maxbitlen);
 788 | 
 789 | /*Compress a buffer with deflate. See RFC 1951. Out buffer must be freed after use.*/
 790 | unsigned lodepng_deflate(unsigned char** out, size_t* outsize,
 791 |                          const unsigned char* in, size_t insize,
 792 |                          const LodePNGCompressSettings* settings);
 793 | 
 794 | #endif /*LODEPNG_COMPILE_ENCODER*/
 795 | #endif /*LODEPNG_COMPILE_ZLIB*/
 796 | 
 797 | #ifdef LODEPNG_COMPILE_DISK
 798 | /*
 799 | Load a file from disk into buffer. The function allocates the out buffer, and
 800 | after usage you should free it.
 801 | out: output parameter, contains pointer to loaded buffer.
 802 | outsize: output parameter, size of the allocated out buffer
 803 | filename: the path to the file to load
 804 | return value: error code (0 means ok)
 805 | */
 806 | unsigned lodepng_load_file(unsigned char** out, size_t* outsize, const char* filename);
 807 | 
 808 | /*
 809 | Save a file from buffer to disk. Warning, if it exists, this function overwrites
 810 | the file without warning!
 811 | buffer: the buffer to write
 812 | buffersize: size of the buffer to write
 813 | filename: the path to the file to save to
 814 | return value: error code (0 means ok)
 815 | */
 816 | unsigned lodepng_save_file(const unsigned char* buffer, size_t buffersize, const char* filename);
 817 | #endif /*LODEPNG_COMPILE_DISK*/
 818 | 
 819 | #ifdef LODEPNG_COMPILE_CPP
 820 | /* The LodePNG C++ wrapper uses std::vectors instead of manually allocated memory buffers. */
 821 | namespace lodepng
 822 | {
 823 | #ifdef LODEPNG_COMPILE_PNG
 824 | class State : public LodePNGState
 825 | {
 826 |   public:
 827 |     State();
 828 |     State(const State& other);
 829 |     virtual ~State();
 830 |     State& operator=(const State& other);
 831 | };
 832 | 
 833 | #ifdef LODEPNG_COMPILE_DECODER
 834 | /* Same as other lodepng::decode, but using a State for more settings and information. */
 835 | unsigned decode(std::vector<unsigned char>& out, unsigned& w, unsigned& h,
 836 |                 State& state,
 837 |                 const unsigned char* in, size_t insize);
 838 | unsigned decode(std::vector<unsigned char>& out, unsigned& w, unsigned& h,
 839 |                 State& state,
 840 |                 const std::vector<unsigned char>& in);
 841 | #endif /*LODEPNG_COMPILE_DECODER*/
 842 | 
 843 | #ifdef LODEPNG_COMPILE_ENCODER
 844 | /* Same as other lodepng::encode, but using a State for more settings and information. */
 845 | unsigned encode(std::vector<unsigned char>& out,
 846 |                 const unsigned char* in, unsigned w, unsigned h,
 847 |                 State& state);
 848 | unsigned encode(std::vector<unsigned char>& out,
 849 |                 const std::vector<unsigned char>& in, unsigned w, unsigned h,
 850 |                 State& state);
 851 | #endif /*LODEPNG_COMPILE_ENCODER*/
 852 | 
 853 | #ifdef LODEPNG_COMPILE_DISK
 854 | /*
 855 | Load a file from disk into an std::vector.
 856 | return value: error code (0 means ok)
 857 | */
 858 | unsigned load_file(std::vector<unsigned char>& buffer, const std::string& filename);
 859 | 
 860 | /*
 861 | Save the binary data in an std::vector to a file on disk. The file is overwritten
 862 | without warning.
 863 | */
 864 | unsigned save_file(const std::vector<unsigned char>& buffer, const std::string& filename);
 865 | #endif /* LODEPNG_COMPILE_DISK */
 866 | #endif /* LODEPNG_COMPILE_PNG */
 867 | 
 868 | #ifdef LODEPNG_COMPILE_ZLIB
 869 | #ifdef LODEPNG_COMPILE_DECODER
 870 | /* Zlib-decompress an unsigned char buffer */
 871 | unsigned decompress(std::vector<unsigned char>& out, const unsigned char* in, size_t insize,
 872 |                     const LodePNGDecompressSettings& settings = lodepng_default_decompress_settings);
 873 | 
 874 | /* Zlib-decompress an std::vector */
 875 | unsigned decompress(std::vector<unsigned char>& out, const std::vector<unsigned char>& in,
 876 |                     const LodePNGDecompressSettings& settings = lodepng_default_decompress_settings);
 877 | #endif /* LODEPNG_COMPILE_DECODER */
 878 | 
 879 | #ifdef LODEPNG_COMPILE_ENCODER
 880 | /* Zlib-compress an unsigned char buffer */
 881 | unsigned compress(std::vector<unsigned char>& out, const unsigned char* in, size_t insize,
 882 |                   const LodePNGCompressSettings& settings = lodepng_default_compress_settings);
 883 | 
 884 | /* Zlib-compress an std::vector */
 885 | unsigned compress(std::vector<unsigned char>& out, const std::vector<unsigned char>& in,
 886 |                   const LodePNGCompressSettings& settings = lodepng_default_compress_settings);
 887 | #endif /* LODEPNG_COMPILE_ENCODER */
 888 | #endif /* LODEPNG_COMPILE_ZLIB */
 889 | } /* namespace lodepng */
 890 | #endif /*LODEPNG_COMPILE_CPP*/
 891 | 
 892 | /*
 893 | TODO:
 894 | [.] test if there are no memory leaks or security exploits - done a lot but needs to be checked often
 895 | [.] check compatibility with various compilers  - done but needs to be redone for every newer version
 896 | [X] converting color to 16-bit per channel types
 897 | [ ] read all public PNG chunk types (but never let the color profile and gamma ones touch RGB values)
 898 | [ ] make sure encoder generates no chunks with size > (2^31)-1
 899 | [ ] partial decoding (stream processing)
 900 | [X] let the "isFullyOpaque" function check color keys and transparent palettes too
 901 | [X] better name for the variables "codes", "codesD", "codelengthcodes", "clcl" and "lldl"
 902 | [ ] don't stop decoding on errors like 69, 57, 58 (make warnings)
 903 | [ ] make warnings like: oob palette, checksum fail, data after iend, wrong/unknown crit chunk, no null terminator in text, ...
 904 | [ ] let the C++ wrapper catch exceptions coming from the standard library and return LodePNG error codes
 905 | [ ] allow user to provide custom color conversion functions, e.g. for premultiplied alpha, padding bits or not, ...
 906 | [ ] allow user to give data (void*) to custom allocator
 907 | */
 908 | 
 909 | #endif /*LODEPNG_H inclusion guard*/
 910 | 
 911 | /*
 912 | LodePNG Documentation
 913 | ---------------------
 914 | 
 915 | 0. table of contents
 916 | --------------------
 917 | 
 918 |   1. about
 919 |    1.1. supported features
 920 |    1.2. features not supported
 921 |   2. C and C++ version
 922 |   3. security
 923 |   4. decoding
 924 |   5. encoding
 925 |   6. color conversions
 926 |     6.1. PNG color types
 927 |     6.2. color conversions
 928 |     6.3. padding bits
 929 |     6.4. A note about 16-bits per channel and endianness
 930 |   7. error values
 931 |   8. chunks and PNG editing
 932 |   9. compiler support
 933 |   10. examples
 934 |    10.1. decoder C++ example
 935 |    10.2. decoder C example
 936 |   11. state settings reference
 937 |   12. changes
 938 |   13. contact information
 939 | 
 940 | 
 941 | 1. about
 942 | --------
 943 | 
 944 | PNG is a file format to store raster images losslessly with good compression,
 945 | supporting different color types and alpha channel.
 946 | 
 947 | LodePNG is a PNG codec according to the Portable Network Graphics (PNG)
 948 | Specification (Second Edition) - W3C Recommendation 10 November 2003.
 949 | 
 950 | The specifications used are:
 951 | 
 952 | *) Portable Network Graphics (PNG) Specification (Second Edition):
 953 |      http://www.w3.org/TR/2003/REC-PNG-20031110
 954 | *) RFC 1950 ZLIB Compressed Data Format version 3.3:
 955 |      http://www.gzip.org/zlib/rfc-zlib.html
 956 | *) RFC 1951 DEFLATE Compressed Data Format Specification ver 1.3:
 957 |      http://www.gzip.org/zlib/rfc-deflate.html
 958 | 
 959 | The most recent version of LodePNG can currently be found at
 960 | http://lodev.org/lodepng/
 961 | 
 962 | LodePNG works both in C (ISO C90) and C++, with a C++ wrapper that adds
 963 | extra functionality.
 964 | 
 965 | LodePNG exists out of two files:
 966 | -lodepng.h: the header file for both C and C++
 967 | -lodepng.c(pp): give it the name lodepng.c or lodepng.cpp (or .cc) depending on your usage
 968 | 
 969 | If you want to start using LodePNG right away without reading this doc, get the
 970 | examples from the LodePNG website to see how to use it in code, or check the
 971 | smaller examples in chapter 13 here.
 972 | 
 973 | LodePNG is simple but only supports the basic requirements. To achieve
 974 | simplicity, the following design choices were made: There are no dependencies
 975 | on any external library. There are functions to decode and encode a PNG with
 976 | a single function call, and extended versions of these functions taking a
 977 | LodePNGState struct allowing to specify or get more information. By default
 978 | the colors of the raw image are always RGB or RGBA, no matter what color type
 979 | the PNG file uses. To read and write files, there are simple functions to
 980 | convert the files to/from buffers in memory.
 981 | 
 982 | This all makes LodePNG suitable for loading textures in games, demos and small
 983 | programs, ... It's less suitable for full fledged image editors, loading PNGs
 984 | over network (it requires all the image data to be available before decoding can
 985 | begin), life-critical systems, ...
 986 | 
 987 | 1.1. supported features
 988 | -----------------------
 989 | 
 990 | The following features are supported by the decoder:
 991 | 
 992 | *) decoding of PNGs with any color type, bit depth and interlace mode, to a 24- or 32-bit color raw image,
 993 |    or the same color type as the PNG
 994 | *) encoding of PNGs, from any raw image to 24- or 32-bit color, or the same color type as the raw image
 995 | *) Adam7 interlace and deinterlace for any color type
 996 | *) loading the image from harddisk or decoding it from a buffer from other sources than harddisk
 997 | *) support for alpha channels, including RGBA color model, translucent palettes and color keying
 998 | *) zlib decompression (inflate)
 999 | *) zlib compression (deflate)
1000 | *) CRC32 and ADLER32 checksums
1001 | *) handling of unknown chunks, allowing making a PNG editor that stores custom and unknown chunks.
1002 | *) the following chunks are supported (generated/interpreted) by both encoder and decoder:
1003 |     IHDR: header information
1004 |     PLTE: color palette
1005 |     IDAT: pixel data
1006 |     IEND: the final chunk
1007 |     tRNS: transparency for palettized images
1008 |     tEXt: textual information
1009 |     zTXt: compressed textual information
1010 |     iTXt: international textual information
1011 |     bKGD: suggested background color
1012 |     pHYs: physical dimensions
1013 |     tIME: modification time
1014 | 
1015 | 1.2. features not supported
1016 | ---------------------------
1017 | 
1018 | The following features are _not_ supported:
1019 | 
1020 | *) some features needed to make a conformant PNG-Editor might be still missing.
1021 | *) partial loading/stream processing. All data must be available and is processed in one call.
1022 | *) The following public chunks are not supported but treated as unknown chunks by LodePNG
1023 |     cHRM, gAMA, iCCP, sRGB, sBIT, hIST, sPLT
1024 |    Some of these are not supported on purpose: LodePNG wants to provide the RGB values
1025 |    stored in the pixels, not values modified by system dependent gamma or color models.
1026 | 
1027 | 
1028 | 2. C and C++ version
1029 | --------------------
1030 | 
1031 | The C version uses buffers allocated with alloc that you need to free()
1032 | yourself. You need to use init and cleanup functions for each struct whenever
1033 | using a struct from the C version to avoid exploits and memory leaks.
1034 | 
1035 | The C++ version has extra functions with std::vectors in the interface and the
1036 | lodepng::State class which is a LodePNGState with constructor and destructor.
1037 | 
1038 | These files work without modification for both C and C++ compilers because all
1039 | the additional C++ code is in "#ifdef __cplusplus" blocks that make C-compilers
1040 | ignore it, and the C code is made to compile both with strict ISO C90 and C++.
1041 | 
1042 | To use the C++ version, you need to rename the source file to lodepng.cpp
1043 | (instead of lodepng.c), and compile it with a C++ compiler.
1044 | 
1045 | To use the C version, you need to rename the source file to lodepng.c (instead
1046 | of lodepng.cpp), and compile it with a C compiler.
1047 | 
1048 | 
1049 | 3. Security
1050 | -----------
1051 | 
1052 | Even if carefully designed, it's always possible that LodePNG contains possible
1053 | exploits. If you discover one, please let me know, and it will be fixed.
1054 | 
1055 | When using LodePNG, care has to be taken with the C version of LodePNG, as well
1056 | as the C-style structs when working with C++. The following conventions are used
1057 | for all C-style structs:
1058 | 
1059 | -if a struct has a corresponding init function, always call the init function when making a new one
1060 | -if a struct has a corresponding cleanup function, call it before the struct disappears to avoid memory leaks
1061 | -if a struct has a corresponding copy function, use the copy function instead of "=".
1062 |  The destination must also be inited already.
1063 | 
1064 | 
1065 | 4. Decoding
1066 | -----------
1067 | 
1068 | Decoding converts a PNG compressed image to a raw pixel buffer.
1069 | 
1070 | Most documentation on using the decoder is at its declarations in the header
1071 | above. For C, simple decoding can be done with functions such as
1072 | lodepng_decode32, and more advanced decoding can be done with the struct
1073 | LodePNGState and lodepng_decode. For C++, all decoding can be done with the
1074 | various lodepng::decode functions, and lodepng::State can be used for advanced
1075 | features.
1076 | 
1077 | When using the LodePNGState, it uses the following fields for decoding:
1078 | *) LodePNGInfo info_png: it stores extra information about the PNG (the input) in here
1079 | *) LodePNGColorMode info_raw: here you can say what color mode of the raw image (the output) you want to get
1080 | *) LodePNGDecoderSettings decoder: you can specify a few extra settings for the decoder to use
1081 | 
1082 | LodePNGInfo info_png
1083 | --------------------
1084 | 
1085 | After decoding, this contains extra information of the PNG image, except the actual
1086 | pixels, width and height because these are already gotten directly from the decoder
1087 | functions.
1088 | 
1089 | It contains for example the original color type of the PNG image, text comments,
1090 | suggested background color, etc... More details about the LodePNGInfo struct are
1091 | at its declaration documentation.
1092 | 
1093 | LodePNGColorMode info_raw
1094 | -------------------------
1095 | 
1096 | When decoding, here you can specify which color type you want
1097 | the resulting raw image to be. If this is different from the colortype of the
1098 | PNG, then the decoder will automatically convert the result. This conversion
1099 | always works, except if you want it to convert a color PNG to greyscale or to
1100 | a palette with missing colors.
1101 | 
1102 | By default, 32-bit color is used for the result.
1103 | 
1104 | LodePNGDecoderSettings decoder
1105 | ------------------------------
1106 | 
1107 | The settings can be used to ignore the errors created by invalid CRC and Adler32
1108 | chunks, and to disable the decoding of tEXt chunks.
1109 | 
1110 | There's also a setting color_convert, true by default. If false, no conversion
1111 | is done, the resulting data will be as it was in the PNG (after decompression)
1112 | and you'll have to puzzle the colors of the pixels together yourself using the
1113 | color type information in the LodePNGInfo.
1114 | 
1115 | 
1116 | 5. Encoding
1117 | -----------
1118 | 
1119 | Encoding converts a raw pixel buffer to a PNG compressed image.
1120 | 
1121 | Most documentation on using the encoder is at its declarations in the header
1122 | above. For C, simple encoding can be done with functions such as
1123 | lodepng_encode32, and more advanced decoding can be done with the struct
1124 | LodePNGState and lodepng_encode. For C++, all encoding can be done with the
1125 | various lodepng::encode functions, and lodepng::State can be used for advanced
1126 | features.
1127 | 
1128 | Like the decoder, the encoder can also give errors. However it gives less errors
1129 | since the encoder input is trusted, the decoder input (a PNG image that could
1130 | be forged by anyone) is not trusted.
1131 | 
1132 | When using the LodePNGState, it uses the following fields for encoding:
1133 | *) LodePNGInfo info_png: here you specify how you want the PNG (the output) to be.
1134 | *) LodePNGColorMode info_raw: here you say what color type of the raw image (the input) has
1135 | *) LodePNGEncoderSettings encoder: you can specify a few settings for the encoder to use
1136 | 
1137 | LodePNGInfo info_png
1138 | --------------------
1139 | 
1140 | When encoding, you use this the opposite way as when decoding: for encoding,
1141 | you fill in the values you want the PNG to have before encoding. By default it's
1142 | not needed to specify a color type for the PNG since it's automatically chosen,
1143 | but it's possible to choose it yourself given the right settings.
1144 | 
1145 | The encoder will not always exactly match the LodePNGInfo struct you give,
1146 | it tries as close as possible. Some things are ignored by the encoder. The
1147 | encoder uses, for example, the following settings from it when applicable:
1148 | colortype and bitdepth, text chunks, time chunk, the color key, the palette, the
1149 | background color, the interlace method, unknown chunks, ...
1150 | 
1151 | When encoding to a PNG with colortype 3, the encoder will generate a PLTE chunk.
1152 | If the palette contains any colors for which the alpha channel is not 255 (so
1153 | there are translucent colors in the palette), it'll add a tRNS chunk.
1154 | 
1155 | LodePNGColorMode info_raw
1156 | -------------------------
1157 | 
1158 | You specify the color type of the raw image that you give to the input here,
1159 | including a possible transparent color key and palette you happen to be using in
1160 | your raw image data.
1161 | 
1162 | By default, 32-bit color is assumed, meaning your input has to be in RGBA
1163 | format with 4 bytes (unsigned chars) per pixel.
1164 | 
1165 | LodePNGEncoderSettings encoder
1166 | ------------------------------
1167 | 
1168 | The following settings are supported (some are in sub-structs):
1169 | *) auto_convert: when this option is enabled, the encoder will
1170 | automatically choose the smallest possible color mode (including color key) that
1171 | can encode the colors of all pixels without information loss.
1172 | *) btype: the block type for LZ77. 0 = uncompressed, 1 = fixed huffman tree,
1173 |    2 = dynamic huffman tree (best compression). Should be 2 for proper
1174 |    compression.
1175 | *) use_lz77: whether or not to use LZ77 for compressed block types. Should be
1176 |    true for proper compression.
1177 | *) windowsize: the window size used by the LZ77 encoder (1 - 32768). Has value
1178 |    2048 by default, but can be set to 32768 for better, but slow, compression.
1179 | *) force_palette: if colortype is 2 or 6, you can make the encoder write a PLTE
1180 |    chunk if force_palette is true. This can used as suggested palette to convert
1181 |    to by viewers that don't support more than 256 colors (if those still exist)
1182 | *) add_id: add text chunk "Encoder: LodePNG <version>" to the image.
1183 | *) text_compression: default 1. If 1, it'll store texts as zTXt instead of tEXt chunks.
1184 |   zTXt chunks use zlib compression on the text. This gives a smaller result on
1185 |   large texts but a larger result on small texts (such as a single program name).
1186 |   It's all tEXt or all zTXt though, there's no separate setting per text yet.
1187 | 
1188 | 
1189 | 6. color conversions
1190 | --------------------
1191 | 
1192 | An important thing to note about LodePNG, is that the color type of the PNG, and
1193 | the color type of the raw image, are completely independent. By default, when
1194 | you decode a PNG, you get the result as a raw image in the color type you want,
1195 | no matter whether the PNG was encoded with a palette, greyscale or RGBA color.
1196 | And if you encode an image, by default LodePNG will automatically choose the PNG
1197 | color type that gives good compression based on the values of colors and amount
1198 | of colors in the image. It can be configured to let you control it instead as
1199 | well, though.
1200 | 
1201 | To be able to do this, LodePNG does conversions from one color mode to another.
1202 | It can convert from almost any color type to any other color type, except the
1203 | following conversions: RGB to greyscale is not supported, and converting to a
1204 | palette when the palette doesn't have a required color is not supported. This is
1205 | not supported on purpose: this is information loss which requires a color
1206 | reduction algorithm that is beyong the scope of a PNG encoder (yes, RGB to grey
1207 | is easy, but there are multiple ways if you want to give some channels more
1208 | weight).
1209 | 
1210 | By default, when decoding, you get the raw image in 32-bit RGBA or 24-bit RGB
1211 | color, no matter what color type the PNG has. And by default when encoding,
1212 | LodePNG automatically picks the best color model for the output PNG, and expects
1213 | the input image to be 32-bit RGBA or 24-bit RGB. So, unless you want to control
1214 | the color format of the images yourself, you can skip this chapter.
1215 | 
1216 | 6.1. PNG color types
1217 | --------------------
1218 | 
1219 | A PNG image can have many color types, ranging from 1-bit color to 64-bit color,
1220 | as well as palettized color modes. After the zlib decompression and unfiltering
1221 | in the PNG image is done, the raw pixel data will have that color type and thus
1222 | a certain amount of bits per pixel. If you want the output raw image after
1223 | decoding to have another color type, a conversion is done by LodePNG.
1224 | 
1225 | The PNG specification gives the following color types:
1226 | 
1227 | 0: greyscale, bit depths 1, 2, 4, 8, 16
1228 | 2: RGB, bit depths 8 and 16
1229 | 3: palette, bit depths 1, 2, 4 and 8
1230 | 4: greyscale with alpha, bit depths 8 and 16
1231 | 6: RGBA, bit depths 8 and 16
1232 | 
1233 | Bit depth is the amount of bits per pixel per color channel. So the total amount
1234 | of bits per pixel is: amount of channels * bitdepth.
1235 | 
1236 | 6.2. color conversions
1237 | ----------------------
1238 | 
1239 | As explained in the sections about the encoder and decoder, you can specify
1240 | color types and bit depths in info_png and info_raw to change the default
1241 | behaviour.
1242 | 
1243 | If, when decoding, you want the raw image to be something else than the default,
1244 | you need to set the color type and bit depth you want in the LodePNGColorMode,
1245 | or the parameters colortype and bitdepth of the simple decoding function.
1246 | 
1247 | If, when encoding, you use another color type than the default in the raw input
1248 | image, you need to specify its color type and bit depth in the LodePNGColorMode
1249 | of the raw image, or use the parameters colortype and bitdepth of the simple
1250 | encoding function.
1251 | 
1252 | If, when encoding, you don't want LodePNG to choose the output PNG color type
1253 | but control it yourself, you need to set auto_convert in the encoder settings
1254 | to false, and specify the color type you want in the LodePNGInfo of the
1255 | encoder (including palette: it can generate a palette if auto_convert is true,
1256 | otherwise not).
1257 | 
1258 | If the input and output color type differ (whether user chosen or auto chosen),
1259 | LodePNG will do a color conversion, which follows the rules below, and may
1260 | sometimes result in an error.
1261 | 
1262 | To avoid some confusion:
1263 | -the decoder converts from PNG to raw image
1264 | -the encoder converts from raw image to PNG
1265 | -the colortype and bitdepth in LodePNGColorMode info_raw, are those of the raw image
1266 | -the colortype and bitdepth in the color field of LodePNGInfo info_png, are those of the PNG
1267 | -when encoding, the color type in LodePNGInfo is ignored if auto_convert
1268 |  is enabled, it is automatically generated instead
1269 | -when decoding, the color type in LodePNGInfo is set by the decoder to that of the original
1270 |  PNG image, but it can be ignored since the raw image has the color type you requested instead
1271 | -if the color type of the LodePNGColorMode and PNG image aren't the same, a conversion
1272 |  between the color types is done if the color types are supported. If it is not
1273 |  supported, an error is returned. If the types are the same, no conversion is done.
1274 | -even though some conversions aren't supported, LodePNG supports loading PNGs from any
1275 |  colortype and saving PNGs to any colortype, sometimes it just requires preparing
1276 |  the raw image correctly before encoding.
1277 | -both encoder and decoder use the same color converter.
1278 | 
1279 | Non supported color conversions:
1280 | -color to greyscale: no error is thrown, but the result will look ugly because
1281 | only the red channel is taken
1282 | -anything to palette when that palette does not have that color in it: in this
1283 | case an error is thrown
1284 | 
1285 | Supported color conversions:
1286 | -anything to 8-bit RGB, 8-bit RGBA, 16-bit RGB, 16-bit RGBA
1287 | -any grey or grey+alpha, to grey or grey+alpha
1288 | -anything to a palette, as long as the palette has the requested colors in it
1289 | -removing alpha channel
1290 | -higher to smaller bitdepth, and vice versa
1291 | 
1292 | If you want no color conversion to be done (e.g. for speed or control):
1293 | -In the encoder, you can make it save a PNG with any color type by giving the
1294 | raw color mode and LodePNGInfo the same color mode, and setting auto_convert to
1295 | false.
1296 | -In the decoder, you can make it store the pixel data in the same color type
1297 | as the PNG has, by setting the color_convert setting to false. Settings in
1298 | info_raw are then ignored.
1299 | 
1300 | The function lodepng_convert does the color conversion. It is available in the
1301 | interface but normally isn't needed since the encoder and decoder already call
1302 | it.
1303 | 
1304 | 6.3. padding bits
1305 | -----------------
1306 | 
1307 | In the PNG file format, if a less than 8-bit per pixel color type is used and the scanlines
1308 | have a bit amount that isn't a multiple of 8, then padding bits are used so that each
1309 | scanline starts at a fresh byte. But that is NOT true for the LodePNG raw input and output.
1310 | The raw input image you give to the encoder, and the raw output image you get from the decoder
1311 | will NOT have these padding bits, e.g. in the case of a 1-bit image with a width
1312 | of 7 pixels, the first pixel of the second scanline will the the 8th bit of the first byte,
1313 | not the first bit of a new byte.
1314 | 
1315 | 6.4. A note about 16-bits per channel and endianness
1316 | ----------------------------------------------------
1317 | 
1318 | LodePNG uses unsigned char arrays for 16-bit per channel colors too, just like
1319 | for any other color format. The 16-bit values are stored in big endian (most
1320 | significant byte first) in these arrays. This is the opposite order of the
1321 | little endian used by x86 CPU's.
1322 | 
1323 | LodePNG always uses big endian because the PNG file format does so internally.
1324 | Conversions to other formats than PNG uses internally are not supported by
1325 | LodePNG on purpose, there are myriads of formats, including endianness of 16-bit
1326 | colors, the order in which you store R, G, B and A, and so on. Supporting and
1327 | converting to/from all that is outside the scope of LodePNG.
1328 | 
1329 | This may mean that, depending on your use case, you may want to convert the big
1330 | endian output of LodePNG to little endian with a for loop. This is certainly not
1331 | always needed, many applications and libraries support big endian 16-bit colors
1332 | anyway, but it means you cannot simply cast the unsigned char* buffer to an
1333 | unsigned short* buffer on x86 CPUs.
1334 | 
1335 | 
1336 | 7. error values
1337 | ---------------
1338 | 
1339 | All functions in LodePNG that return an error code, return 0 if everything went
1340 | OK, or a non-zero code if there was an error.
1341 | 
1342 | The meaning of the LodePNG error values can be retrieved with the function
1343 | lodepng_error_text: given the numerical error code, it returns a description
1344 | of the error in English as a string.
1345 | 
1346 | Check the implementation of lodepng_error_text to see the meaning of each code.
1347 | 
1348 | 
1349 | 8. chunks and PNG editing
1350 | -------------------------
1351 | 
1352 | If you want to add extra chunks to a PNG you encode, or use LodePNG for a PNG
1353 | editor that should follow the rules about handling of unknown chunks, or if your
1354 | program is able to read other types of chunks than the ones handled by LodePNG,
1355 | then that's possible with the chunk functions of LodePNG.
1356 | 
1357 | A PNG chunk has the following layout:
1358 | 
1359 | 4 bytes length
1360 | 4 bytes type name
1361 | length bytes data
1362 | 4 bytes CRC
1363 | 
1364 | 8.1. iterating through chunks
1365 | -----------------------------
1366 | 
1367 | If you have a buffer containing the PNG image data, then the first chunk (the
1368 | IHDR chunk) starts at byte number 8 of that buffer. The first 8 bytes are the
1369 | signature of the PNG and are not part of a chunk. But if you start at byte 8
1370 | then you have a chunk, and can check the following things of it.
1371 | 
1372 | NOTE: none of these functions check for memory buffer boundaries. To avoid
1373 | exploits, always make sure the buffer contains all the data of the chunks.
1374 | When using lodepng_chunk_next, make sure the returned value is within the
1375 | allocated memory.
1376 | 
1377 | unsigned lodepng_chunk_length(const unsigned char* chunk):
1378 | 
1379 | Get the length of the chunk's data. The total chunk length is this length + 12.
1380 | 
1381 | void lodepng_chunk_type(char type[5], const unsigned char* chunk):
1382 | unsigned char lodepng_chunk_type_equals(const unsigned char* chunk, const char* type):
1383 | 
1384 | Get the type of the chunk or compare if it's a certain type
1385 | 
1386 | unsigned char lodepng_chunk_critical(const unsigned char* chunk):
1387 | unsigned char lodepng_chunk_private(const unsigned char* chunk):
1388 | unsigned char lodepng_chunk_safetocopy(const unsigned char* chunk):
1389 | 
1390 | Check if the chunk is critical in the PNG standard (only IHDR, PLTE, IDAT and IEND are).
1391 | Check if the chunk is private (public chunks are part of the standard, private ones not).
1392 | Check if the chunk is safe to copy. If it's not, then, when modifying data in a critical
1393 | chunk, unsafe to copy chunks of the old image may NOT be saved in the new one if your
1394 | program doesn't handle that type of unknown chunk.
1395 | 
1396 | unsigned char* lodepng_chunk_data(unsigned char* chunk):
1397 | const unsigned char* lodepng_chunk_data_const(const unsigned char* chunk):
1398 | 
1399 | Get a pointer to the start of the data of the chunk.
1400 | 
1401 | unsigned lodepng_chunk_check_crc(const unsigned char* chunk):
1402 | void lodepng_chunk_generate_crc(unsigned char* chunk):
1403 | 
1404 | Check if the crc is correct or generate a correct one.
1405 | 
1406 | unsigned char* lodepng_chunk_next(unsigned char* chunk):
1407 | const unsigned char* lodepng_chunk_next_const(const unsigned char* chunk):
1408 | 
1409 | Iterate to the next chunk. This works if you have a buffer with consecutive chunks. Note that these
1410 | functions do no boundary checking of the allocated data whatsoever, so make sure there is enough
1411 | data available in the buffer to be able to go to the next chunk.
1412 | 
1413 | unsigned lodepng_chunk_append(unsigned char** out, size_t* outlength, const unsigned char* chunk):
1414 | unsigned lodepng_chunk_create(unsigned char** out, size_t* outlength, unsigned length,
1415 |                               const char* type, const unsigned char* data):
1416 | 
1417 | These functions are used to create new chunks that are appended to the data in *out that has
1418 | length *outlength. The append function appends an existing chunk to the new data. The create
1419 | function creates a new chunk with the given parameters and appends it. Type is the 4-letter
1420 | name of the chunk.
1421 | 
1422 | 8.2. chunks in info_png
1423 | -----------------------
1424 | 
1425 | The LodePNGInfo struct contains fields with the unknown chunk in it. It has 3
1426 | buffers (each with size) to contain 3 types of unknown chunks:
1427 | the ones that come before the PLTE chunk, the ones that come between the PLTE
1428 | and the IDAT chunks, and the ones that come after the IDAT chunks.
1429 | It's necessary to make the distionction between these 3 cases because the PNG
1430 | standard forces to keep the ordering of unknown chunks compared to the critical
1431 | chunks, but does not force any other ordering rules.
1432 | 
1433 | info_png.unknown_chunks_data[0] is the chunks before PLTE
1434 | info_png.unknown_chunks_data[1] is the chunks after PLTE, before IDAT
1435 | info_png.unknown_chunks_data[2] is the chunks after IDAT
1436 | 
1437 | The chunks in these 3 buffers can be iterated through and read by using the same
1438 | way described in the previous subchapter.
1439 | 
1440 | When using the decoder to decode a PNG, you can make it store all unknown chunks
1441 | if you set the option settings.remember_unknown_chunks to 1. By default, this
1442 | option is off (0).
1443 | 
1444 | The encoder will always encode unknown chunks that are stored in the info_png.
1445 | If you need it to add a particular chunk that isn't known by LodePNG, you can
1446 | use lodepng_chunk_append or lodepng_chunk_create to the chunk data in
1447 | info_png.unknown_chunks_data[x].
1448 | 
1449 | Chunks that are known by LodePNG should not be added in that way. E.g. to make
1450 | LodePNG add a bKGD chunk, set background_defined to true and add the correct
1451 | parameters there instead.
1452 | 
1453 | 
1454 | 9. compiler support
1455 | -------------------
1456 | 
1457 | No libraries other than the current standard C library are needed to compile
1458 | LodePNG. For the C++ version, only the standard C++ library is needed on top.
1459 | Add the files lodepng.c(pp) and lodepng.h to your project, include
1460 | lodepng.h where needed, and your program can read/write PNG files.
1461 | 
1462 | It is compatible with C90 and up, and C++03 and up.
1463 | 
1464 | If performance is important, use optimization when compiling! For both the
1465 | encoder and decoder, this makes a large difference.
1466 | 
1467 | Make sure that LodePNG is compiled with the same compiler of the same version
1468 | and with the same settings as the rest of the program, or the interfaces with
1469 | std::vectors and std::strings in C++ can be incompatible.
1470 | 
1471 | CHAR_BITS must be 8 or higher, because LodePNG uses unsigned chars for octets.
1472 | 
1473 | *) gcc and g++
1474 | 
1475 | LodePNG is developed in gcc so this compiler is natively supported. It gives no
1476 | warnings with compiler options "-Wall -Wextra -pedantic -ansi", with gcc and g++
1477 | version 4.7.1 on Linux, 32-bit and 64-bit.
1478 | 
1479 | *) Clang
1480 | 
1481 | Fully supported and warning-free.
1482 | 
1483 | *) Mingw
1484 | 
1485 | The Mingw compiler (a port of gcc for Windows) should be fully supported by
1486 | LodePNG.
1487 | 
1488 | *) Visual Studio and Visual C++ Express Edition
1489 | 
1490 | LodePNG should be warning-free with warning level W4. Two warnings were disabled
1491 | with pragmas though: warning 4244 about implicit conversions, and warning 4996
1492 | where it wants to use a non-standard function fopen_s instead of the standard C
1493 | fopen.
1494 | 
1495 | Visual Studio may want "stdafx.h" files to be included in each source file and
1496 | give an error "unexpected end of file while looking for precompiled header".
1497 | This is not standard C++ and will not be added to the stock LodePNG. You can
1498 | disable it for lodepng.cpp only by right clicking it, Properties, C/C++,
1499 | Precompiled Headers, and set it to Not Using Precompiled Headers there.
1500 | 
1501 | NOTE: Modern versions of VS should be fully supported, but old versions, e.g.
1502 | VS6, are not guaranteed to work.
1503 | 
1504 | *) Compilers on Macintosh
1505 | 
1506 | LodePNG has been reported to work both with gcc and LLVM for Macintosh, both for
1507 | C and C++.
1508 | 
1509 | *) Other Compilers
1510 | 
1511 | If you encounter problems on any compilers, feel free to let me know and I may
1512 | try to fix it if the compiler is modern and standards complient.
1513 | 
1514 | 
1515 | 10. examples
1516 | ------------
1517 | 
1518 | This decoder example shows the most basic usage of LodePNG. More complex
1519 | examples can be found on the LodePNG website.
1520 | 
1521 | 10.1. decoder C++ example
1522 | -------------------------
1523 | 
1524 | #include "lodepng.h"
1525 | #include <iostream>
1526 | 
1527 | int main(int argc, char *argv[])
1528 | {
1529 |   const char* filename = argc > 1 ? argv[1] : "test.png";
1530 | 
1531 |   //load and decode
1532 |   std::vector<unsigned char> image;
1533 |   unsigned width, height;
1534 |   unsigned error = lodepng::decode(image, width, height, filename);
1535 | 
1536 |   //if there's an error, display it
1537 |   if(error) std::cout << "decoder error " << error << ": " << lodepng_error_text(error) << std::endl;
1538 | 
1539 |   //the pixels are now in the vector "image", 4 bytes per pixel, ordered RGBARGBA..., use it as texture, draw it, ...
1540 | }
1541 | 
1542 | 10.2. decoder C example
1543 | -----------------------
1544 | 
1545 | #include "lodepng.h"
1546 | 
1547 | int main(int argc, char *argv[])
1548 | {
1549 |   unsigned error;
1550 |   unsigned char* image;
1551 |   size_t width, height;
1552 |   const char* filename = argc > 1 ? argv[1] : "test.png";
1553 | 
1554 |   error = lodepng_decode32_file(&image, &width, &height, filename);
1555 | 
1556 |   if(error) printf("decoder error %u: %s\n", error, lodepng_error_text(error));
1557 | 
1558 |   / * use image here * /
1559 | 
1560 |   free(image);
1561 |   return 0;
1562 | }
1563 | 
1564 | 11. state settings reference
1565 | ----------------------------
1566 | 
1567 | A quick reference of some settings to set on the LodePNGState
1568 | 
1569 | For decoding:
1570 | 
1571 | state.decoder.zlibsettings.ignore_adler32: ignore ADLER32 checksums
1572 | state.decoder.zlibsettings.custom_...: use custom inflate function
1573 | state.decoder.ignore_crc: ignore CRC checksums
1574 | state.decoder.ignore_critical: ignore unknown critical chunks
1575 | state.decoder.ignore_end: ignore missing IEND chunk. May fail if this corruption causes other errors
1576 | state.decoder.color_convert: convert internal PNG color to chosen one
1577 | state.decoder.read_text_chunks: whether to read in text metadata chunks
1578 | state.decoder.remember_unknown_chunks: whether to read in unknown chunks
1579 | state.info_raw.colortype: desired color type for decoded image
1580 | state.info_raw.bitdepth: desired bit depth for decoded image
1581 | state.info_raw....: more color settings, see struct LodePNGColorMode
1582 | state.info_png....: no settings for decoder but ouput, see struct LodePNGInfo
1583 | 
1584 | For encoding:
1585 | 
1586 | state.encoder.zlibsettings.btype: disable compression by setting it to 0
1587 | state.encoder.zlibsettings.use_lz77: use LZ77 in compression
1588 | state.encoder.zlibsettings.windowsize: tweak LZ77 windowsize
1589 | state.encoder.zlibsettings.minmatch: tweak min LZ77 length to match
1590 | state.encoder.zlibsettings.nicematch: tweak LZ77 match where to stop searching
1591 | state.encoder.zlibsettings.lazymatching: try one more LZ77 matching
1592 | state.encoder.zlibsettings.custom_...: use custom deflate function
1593 | state.encoder.auto_convert: choose optimal PNG color type, if 0 uses info_png
1594 | state.encoder.filter_palette_zero: PNG filter strategy for palette
1595 | state.encoder.filter_strategy: PNG filter strategy to encode with
1596 | state.encoder.force_palette: add palette even if not encoding to one
1597 | state.encoder.add_id: add LodePNG identifier and version as a text chunk
1598 | state.encoder.text_compression: use compressed text chunks for metadata
1599 | state.info_raw.colortype: color type of raw input image you provide
1600 | state.info_raw.bitdepth: bit depth of raw input image you provide
1601 | state.info_raw: more color settings, see struct LodePNGColorMode
1602 | state.info_png.color.colortype: desired color type if auto_convert is false
1603 | state.info_png.color.bitdepth: desired bit depth if auto_convert is false
1604 | state.info_png.color....: more color settings, see struct LodePNGColorMode
1605 | state.info_png....: more PNG related settings, see struct LodePNGInfo
1606 | 
1607 | 
1608 | 12. changes
1609 | -----------
1610 | 
1611 | The version number of LodePNG is the date of the change given in the format
1612 | yyyymmdd.
1613 | 
1614 | Some changes aren't backwards compatible. Those are indicated with a (!)
1615 | symbol.
1616 | 
1617 | *) 14 jan 2018: allow optionally ignoring a few more recoverable errors
1618 | *) 17 sep 2017: fix memory leak for some encoder input error cases
1619 | *) 27 nov 2016: grey+alpha auto color model detection bugfix
1620 | *) 18 apr 2016: Changed qsort to custom stable sort (for platforms w/o qsort).
1621 | *) 09 apr 2016: Fixed colorkey usage detection, and better file loading (within
1622 |    the limits of pure C90).
1623 | *) 08 dec 2015: Made load_file function return error if file can't be opened.
1624 | *) 24 okt 2015: Bugfix with decoding to palette output.
1625 | *) 18 apr 2015: Boundary PM instead of just package-merge for faster encoding.
1626 | *) 23 aug 2014: Reduced needless memory usage of decoder.
1627 | *) 28 jun 2014: Removed fix_png setting, always support palette OOB for
1628 |     simplicity. Made ColorProfile public.
1629 | *) 09 jun 2014: Faster encoder by fixing hash bug and more zeros optimization.
1630 | *) 22 dec 2013: Power of two windowsize required for optimization.
1631 | *) 15 apr 2013: Fixed bug with LAC_ALPHA and color key.
1632 | *) 25 mar 2013: Added an optional feature to ignore some PNG errors (fix_png).
1633 | *) 11 mar 2013 (!): Bugfix with custom free. Changed from "my" to "lodepng_"
1634 |     prefix for the custom allocators and made it possible with a new #define to
1635 |     use custom ones in your project without needing to change lodepng's code.
1636 | *) 28 jan 2013: Bugfix with color key.
1637 | *) 27 okt 2012: Tweaks in text chunk keyword length error handling.
1638 | *) 8 okt 2012 (!): Added new filter strategy (entropy) and new auto color mode.
1639 |     (no palette). Better deflate tree encoding. New compression tweak settings.
1640 |     Faster color conversions while decoding. Some internal cleanups.
1641 | *) 23 sep 2012: Reduced warnings in Visual Studio a little bit.
1642 | *) 1 sep 2012 (!): Removed #define's for giving custom (de)compression functions
1643 |     and made it work with function pointers instead.
1644 | *) 23 jun 2012: Added more filter strategies. Made it easier to use custom alloc
1645 |     and free functions and toggle #defines from compiler flags. Small fixes.
1646 | *) 6 may 2012 (!): Made plugging in custom zlib/deflate functions more flexible.
1647 | *) 22 apr 2012 (!): Made interface more consistent, renaming a lot. Removed
1648 |     redundant C++ codec classes. Reduced amount of structs. Everything changed,
1649 |     but it is cleaner now imho and functionality remains the same. Also fixed
1650 |     several bugs and shrunk the implementation code. Made new samples.
1651 | *) 6 nov 2011 (!): By default, the encoder now automatically chooses the best
1652 |     PNG color model and bit depth, based on the amount and type of colors of the
1653 |     raw image. For this, autoLeaveOutAlphaChannel replaced by auto_choose_color.
1654 | *) 9 okt 2011: simpler hash chain implementation for the encoder.
1655 | *) 8 sep 2011: lz77 encoder lazy matching instead of greedy matching.
1656 | *) 23 aug 2011: tweaked the zlib compression parameters after benchmarking.
1657 |     A bug with the PNG filtertype heuristic was fixed, so that it chooses much
1658 |     better ones (it's quite significant). A setting to do an experimental, slow,
1659 |     brute force search for PNG filter types is added.
1660 | *) 17 aug 2011 (!): changed some C zlib related function names.
1661 | *) 16 aug 2011: made the code less wide (max 120 characters per line).
1662 | *) 17 apr 2011: code cleanup. Bugfixes. Convert low to 16-bit per sample colors.
1663 | *) 21 feb 2011: fixed compiling for C90. Fixed compiling with sections disabled.
1664 | *) 11 dec 2010: encoding is made faster, based on suggestion by Peter Eastman
1665 |     to optimize long sequences of zeros.
1666 | *) 13 nov 2010: added LodePNG_InfoColor_hasPaletteAlpha and
1667 |     LodePNG_InfoColor_canHaveAlpha functions for convenience.
1668 | *) 7 nov 2010: added LodePNG_error_text function to get error code description.
1669 | *) 30 okt 2010: made decoding slightly faster
1670 | *) 26 okt 2010: (!) changed some C function and struct names (more consistent).
1671 |      Reorganized the documentation and the declaration order in the header.
1672 | *) 08 aug 2010: only changed some comments and external samples.
1673 | *) 05 jul 2010: fixed bug thanks to warnings in the new gcc version.
1674 | *) 14 mar 2010: fixed bug where too much memory was allocated for char buffers.
1675 | *) 02 sep 2008: fixed bug where it could create empty tree that linux apps could
1676 |     read by ignoring the problem but windows apps couldn't.
1677 | *) 06 jun 2008: added more error checks for out of memory cases.
1678 | *) 26 apr 2008: added a few more checks here and there to ensure more safety.
1679 | *) 06 mar 2008: crash with encoding of strings fixed
1680 | *) 02 feb 2008: support for international text chunks added (iTXt)
1681 | *) 23 jan 2008: small cleanups, and #defines to divide code in sections
1682 | *) 20 jan 2008: support for unknown chunks allowing using LodePNG for an editor.
1683 | *) 18 jan 2008: support for tIME and pHYs chunks added to encoder and decoder.
1684 | *) 17 jan 2008: ability to encode and decode compressed zTXt chunks added
1685 |     Also various fixes, such as in the deflate and the padding bits code.
1686 | *) 13 jan 2008: Added ability to encode Adam7-interlaced images. Improved
1687 |     filtering code of encoder.
1688 | *) 07 jan 2008: (!) changed LodePNG to use ISO C90 instead of C++. A
1689 |     C++ wrapper around this provides an interface almost identical to before.
1690 |     Having LodePNG be pure ISO C90 makes it more portable. The C and C++ code
1691 |     are together in these files but it works both for C and C++ compilers.
1692 | *) 29 dec 2007: (!) changed most integer types to unsigned int + other tweaks
1693 | *) 30 aug 2007: bug fixed which makes this Borland C++ compatible
1694 | *) 09 aug 2007: some VS2005 warnings removed again
1695 | *) 21 jul 2007: deflate code placed in new namespace separate from zlib code
1696 | *) 08 jun 2007: fixed bug with 2- and 4-bit color, and small interlaced images
1697 | *) 04 jun 2007: improved support for Visual Studio 2005: crash with accessing
1698 |     invalid std::vector element [0] fixed, and level 3 and 4 warnings removed
1699 | *) 02 jun 2007: made the encoder add a tag with version by default
1700 | *) 27 may 2007: zlib and png code separated (but still in the same file),
1701 |     simple encoder/decoder functions added for more simple usage cases
1702 | *) 19 may 2007: minor fixes, some code cleaning, new error added (error 69),
1703 |     moved some examples from here to lodepng_examples.cpp
1704 | *) 12 may 2007: palette decoding bug fixed
1705 | *) 24 apr 2007: changed the license from BSD to the zlib license
1706 | *) 11 mar 2007: very simple addition: ability to encode bKGD chunks.
1707 | *) 04 mar 2007: (!) tEXt chunk related fixes, and support for encoding
1708 |     palettized PNG images. Plus little interface change with palette and texts.
1709 | *) 03 mar 2007: Made it encode dynamic Huffman shorter with repeat codes.
1710 |     Fixed a bug where the end code of a block had length 0 in the Huffman tree.
1711 | *) 26 feb 2007: Huffman compression with dynamic trees (BTYPE 2) now implemented
1712 |     and supported by the encoder, resulting in smaller PNGs at the output.
1713 | *) 27 jan 2007: Made the Adler-32 test faster so that a timewaste is gone.
1714 | *) 24 jan 2007: gave encoder an error interface. Added color conversion from any
1715 |     greyscale type to 8-bit greyscale with or without alpha.
1716 | *) 21 jan 2007: (!) Totally changed the interface. It allows more color types
1717 |     to convert to and is more uniform. See the manual for how it works now.
1718 | *) 07 jan 2007: Some cleanup & fixes, and a few changes over the last days:
1719 |     encode/decode custom tEXt chunks, separate classes for zlib & deflate, and
1720 |     at last made the decoder give errors for incorrect Adler32 or Crc.
1721 | *) 01 jan 2007: Fixed bug with encoding PNGs with less than 8 bits per channel.
1722 | *) 29 dec 2006: Added support for encoding images without alpha channel, and
1723 |     cleaned out code as well as making certain parts faster.
1724 | *) 28 dec 2006: Added "Settings" to the encoder.
1725 | *) 26 dec 2006: The encoder now does LZ77 encoding and produces much smaller files now.
1726 |     Removed some code duplication in the decoder. Fixed little bug in an example.
1727 | *) 09 dec 2006: (!) Placed output parameters of public functions as first parameter.
1728 |     Fixed a bug of the decoder with 16-bit per color.
1729 | *) 15 okt 2006: Changed documentation structure
1730 | *) 09 okt 2006: Encoder class added. It encodes a valid PNG image from the
1731 |     given image buffer, however for now it's not compressed.
1732 | *) 08 sep 2006: (!) Changed to interface with a Decoder class
1733 | *) 30 jul 2006: (!) LodePNG_InfoPng , width and height are now retrieved in different
1734 |     way. Renamed decodePNG to decodePNGGeneric.
1735 | *) 29 jul 2006: (!) Changed the interface: image info is now returned as a
1736 |     struct of type LodePNG::LodePNG_Info, instead of a vector, which was a bit clumsy.
1737 | *) 28 jul 2006: Cleaned the code and added new error checks.
1738 |     Corrected terminology "deflate" into "inflate".
1739 | *) 23 jun 2006: Added SDL example in the documentation in the header, this
1740 |     example allows easy debugging by displaying the PNG and its transparency.
1741 | *) 22 jun 2006: (!) Changed way to obtain error value. Added
1742 |     loadFile function for convenience. Made decodePNG32 faster.
1743 | *) 21 jun 2006: (!) Changed type of info vector to unsigned.
1744 |     Changed position of palette in info vector. Fixed an important bug that
1745 |     happened on PNGs with an uncompressed block.
1746 | *) 16 jun 2006: Internally changed unsigned into unsigned where
1747 |     needed, and performed some optimizations.
1748 | *) 07 jun 2006: (!) Renamed functions to decodePNG and placed them
1749 |     in LodePNG namespace. Changed the order of the parameters. Rewrote the
1750 |     documentation in the header. Renamed files to lodepng.cpp and lodepng.h
1751 | *) 22 apr 2006: Optimized and improved some code
1752 | *) 07 sep 2005: (!) Changed to std::vector interface
1753 | *) 12 aug 2005: Initial release (C++, decoder only)
1754 | 
1755 | 
1756 | 13. contact information
1757 | -----------------------
1758 | 
1759 | Feel free to contact me with suggestions, problems, comments, ... concerning
1760 | LodePNG. If you encounter a PNG image that doesn't work properly with this
1761 | decoder, feel free to send it and I'll use it to find and fix the problem.
1762 | 
1763 | My email address is (puzzle the account and domain together with an @ symbol):
1764 | Domain: gmail dot com.
1765 | Account: lode dot vandevenne.
1766 | 
1767 | 
1768 | Copyright (c) 2005-2017 Lode Vandevenne
1769 | */
1770 | 


--------------------------------------------------------------------------------