├── .github └── FUNDING.yml ├── LICENSE ├── README.md ├── doc ├── crop.png └── outline.png ├── example ├── 25535163354.jpg ├── 26939476984.jpg ├── 27417460620.jpg ├── 28411051634.jpg ├── 28922730122.jpg ├── 28930160605.jpg ├── 6833735316.jpg ├── 8527042251.jpg ├── example.go └── output.html ├── example2 └── example2.go ├── imgprep.go └── kmeans.go /.github/FUNDING.yml: -------------------------------------------------------------------------------- 1 | ko_fi: supportadev 2 | -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | Copyright (c) 2016 Carl Asman. All rights reserved. 2 | 3 | Redistribution and use in source and binary forms, with or without 4 | modification, are permitted provided that the following conditions are 5 | met: 6 | 7 | * Redistributions of source code must retain the above copyright 8 | notice, this list of conditions and the following disclaimer. 9 | * Redistributions in binary form must reproduce the above 10 | copyright notice, this list of conditions and the following disclaimer 11 | in the documentation and/or other materials provided with the 12 | distribution. 13 | * Neither the name of the copyright holder nor the names of its 14 | contributors may be used to endorse or promote products derived from 15 | this software without specific prior written permission. 16 | 17 | THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 18 | "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 19 | LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 20 | A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 21 | OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 22 | SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 23 | LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 24 | DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 25 | THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 26 | (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 27 | OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 28 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | [![Go Report Card](https://goreportcard.com/badge/github.com/EdlinOrg/prominentcolor)](https://goreportcard.com/report/github.com/EdlinOrg/prominentcolor) 2 | [![GoDoc](https://godoc.org/github.com/EdlinOrg/prominentcolor?status.svg)](https://godoc.org/github.com/EdlinOrg/prominentcolor) 3 | 4 | # prominentcolor 5 | 6 | ## Find the K most dominant colors in an image 7 | 8 | The `Kmeans` function returns the K most dominant colors in the image, ordered in the order of dominance. 9 | 10 | To execute it with the default values, call 11 | 12 | `func Kmeans(orgimg image.Image) (centroids []ColorItem)` 13 | 14 | which takes an image and returns the K dominant colors, sorted by the most frequent one first. 15 | 16 | It uses Kmeans++ to pick K initial centroid values, and goes through all pixels to re-calculate the centroids used. 17 | 18 | As default function `Kmeans` uses these settings: 19 | * K=3 20 | * crops the center of the image before resizing (removing 25% on all sides) 21 | * resizes to 80 pixels 22 | * uses Kmeans++ 23 | * uses median value to find the color for the centroid 24 | * mask out white, black or green backgrounds 25 | 26 | To have more control, call `KmeansWithArgs` or `KmeansWithAll`. 27 | Below are the parameters that can be tweaked when calling those functions. 28 | 29 | ## K 30 | As default it has got K=3. 31 | 32 | If set to high, it will get too detailed and would separate nuances of the same color in different centroids. 33 | 34 | ## Resizing 35 | As default it resizes the image to 80 pixels wide (and whatever height to preserve aspect ratio). 36 | 37 | The higher value, the more time it will take to process since it goes through all pixels. 38 | 39 | ## Arguments 40 | 41 | ### `ArgumentSeedRandom` : Kmeans++ vs Random 42 | As default it uses Kmeans++. 43 | 44 | Kmeans++ will take K points that are as far away from each other as possible, 45 | to avoid that the points might be too close to each other and really could be in the same cluster. 46 | Hence the initial step takes slightly longer than just randomly picking the initial K starting points. 47 | 48 | ### `ArgumentAverageMean` : Median vs mean for picking color 49 | As default it uses median. 50 | 51 | When the colors are being lumped together in the K clusters, it can pick the _mean_ value, meaning 52 | adding all values together and dividing by the number of colors in that cluster. 53 | This will make the centroid color to be close to the color of the majority of the pixels in that cluster. 54 | Median will take the median value, i.e. just take the one in the middle of all colors in the cluster. 55 | 56 | ### `ArgumentNoCropping` : Crop to center of image vs not cropping 57 | 58 | As default, it crops the center of the image (removing 25% on all sides). 59 | 60 | The theory being that the most relevant area in the image is in the middle, 61 | so even if the most dominant color in that area is not the most dominant color in the whole image, it might be what the user percieve as most dominant. 62 | 63 | The image below contains mostly green color, but since the flower is in the center of the image, we might be interested in perceiving that as the dominant color. When cropping (default) it finds pink to be most dominant, without cropping (by setting `ArgumentNoCropping`), green is most dominant. 64 | 65 | ![Using cropCenter](doc/crop.png) 66 | 67 | ### `ArgumentLAB` : RGB vs LAB 68 | 69 | As default it uses RGB. 70 | 71 | LAB is experimental atm, hence RGB is default. 72 | 73 | ### `ArgumentDebugImage` : Save temporary image 74 | 75 | Saves an image in `/tmp/` where the pixels that have been masked out are colored pink. 76 | Useful when modifying the values of the masks, so you can observe the result. 77 | 78 | ## Masking; removing background colours 79 | 80 | `GetDefaultMasks` is the function containing the masks used as default, they can be used as a starting point 81 | when passing other masks to the function. As default it filters white, black or green backgrounds. 82 | 83 | To handle items that are shot against white/black/green background ("isolated" objects / clipart / green screen images), 84 | the image is pre-processed to disregard the white/black/green background: 85 | If the four corners are in that same color (or close to it), the code will take those as starting points for the areas to be removed. 86 | 87 | In the image below, it removes much of the white (the pink pixels are the pixels that have been removed). 88 | By removing those areas, "white" will have less of a chance of becoming the dominant color. 89 | 90 | ![Ignoring backgrounds](doc/outline.png) 91 | 92 | ## Sample code 93 | 94 | See 95 | [example/example.go](example/example.go) 96 | and 97 | [example2/example2.go](example2/example2.go) 98 | for sample calls with different parameters. 99 | 100 | The sample images in 101 | [example/](example/) 102 | comes from flickr and are all Public Domain https://creativecommons.org/publicdomain/zero/1.0/ 103 | 104 | The images used: 105 | * https://www.flickr.com/photos/65720474@N03/8527042251/ 106 | * https://www.flickr.com/photos/isasza/26939476984/ 107 | * https://www.flickr.com/photos/janosvirag/27417460620/ 108 | * https://www.flickr.com/photos/janosvirag/28922730122/ 109 | * https://www.flickr.com/photos/mathiasappel/25535163354/ 110 | * https://www.flickr.com/photos/mathiasappel/28930160605/ 111 | * https://www.flickr.com/photos/pasukaru76/6833735316/ 112 | * https://www.flickr.com/photos/sloalan/28411051634/ 113 | 114 | ## Author 115 | 116 | Carl Asman (www.edlin.org) 117 | 118 | ## BSD License 119 | 120 | See [LICENSE](LICENSE) 121 | -------------------------------------------------------------------------------- /doc/crop.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/EdlinOrg/prominentcolor/0cadac69bd03e873f418712f44574837d63d7755/doc/crop.png -------------------------------------------------------------------------------- /doc/outline.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/EdlinOrg/prominentcolor/0cadac69bd03e873f418712f44574837d63d7755/doc/outline.png -------------------------------------------------------------------------------- /example/25535163354.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/EdlinOrg/prominentcolor/0cadac69bd03e873f418712f44574837d63d7755/example/25535163354.jpg -------------------------------------------------------------------------------- /example/26939476984.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/EdlinOrg/prominentcolor/0cadac69bd03e873f418712f44574837d63d7755/example/26939476984.jpg -------------------------------------------------------------------------------- /example/27417460620.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/EdlinOrg/prominentcolor/0cadac69bd03e873f418712f44574837d63d7755/example/27417460620.jpg -------------------------------------------------------------------------------- /example/28411051634.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/EdlinOrg/prominentcolor/0cadac69bd03e873f418712f44574837d63d7755/example/28411051634.jpg -------------------------------------------------------------------------------- /example/28922730122.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/EdlinOrg/prominentcolor/0cadac69bd03e873f418712f44574837d63d7755/example/28922730122.jpg -------------------------------------------------------------------------------- /example/28930160605.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/EdlinOrg/prominentcolor/0cadac69bd03e873f418712f44574837d63d7755/example/28930160605.jpg -------------------------------------------------------------------------------- /example/6833735316.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/EdlinOrg/prominentcolor/0cadac69bd03e873f418712f44574837d63d7755/example/6833735316.jpg -------------------------------------------------------------------------------- /example/8527042251.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/EdlinOrg/prominentcolor/0cadac69bd03e873f418712f44574837d63d7755/example/8527042251.jpg -------------------------------------------------------------------------------- /example/example.go: -------------------------------------------------------------------------------- 1 | package main 2 | 3 | import ( 4 | "fmt" 5 | "image" 6 | _ "image/jpeg" 7 | "io/ioutil" 8 | "log" 9 | "os" 10 | "strings" 11 | 12 | prominentcolor ".." 13 | ) 14 | 15 | func loadImage(fileInput string) (image.Image, error) { 16 | f, err := os.Open(fileInput) 17 | defer f.Close() 18 | if err != nil { 19 | log.Println("File not found:", fileInput) 20 | return nil, err 21 | } 22 | img, _, err := image.Decode(f) 23 | if err != nil { 24 | return nil, err 25 | } 26 | 27 | return img, nil 28 | } 29 | 30 | func outputColorRange(colorRange []prominentcolor.ColorItem) string { 31 | var buff strings.Builder 32 | buff.WriteString("") 33 | for _, color := range colorRange { 34 | buff.WriteString(fmt.Sprintf("", color.AsString(), color.AsString(), color.Cnt)) 35 | } 36 | buff.WriteString("
#%s %d
") 37 | return buff.String() 38 | } 39 | 40 | func outputTitle(str string) string { 41 | return "

" + str + "

" 42 | } 43 | 44 | func processBatch(k int, bitarr []int, img image.Image) string { 45 | var buff strings.Builder 46 | 47 | prefix := fmt.Sprintf("K=%d, ", k) 48 | resizeSize := uint(prominentcolor.DefaultSize) 49 | bgmasks := prominentcolor.GetDefaultMasks() 50 | 51 | for i := 0; i < len(bitarr); i++ { 52 | res, err := prominentcolor.KmeansWithAll(k, img, bitarr[i], resizeSize, bgmasks) 53 | if err != nil { 54 | log.Println(err) 55 | continue 56 | } 57 | buff.WriteString(outputTitle(prefix + bitInfo(bitarr[i]))) 58 | buff.WriteString(outputColorRange(res)) 59 | } 60 | 61 | return buff.String() 62 | } 63 | 64 | func bitInfo(bits int) string { 65 | list := make([]string, 0, 4) 66 | // random seed or Kmeans++ 67 | if prominentcolor.IsBitSet(bits, prominentcolor.ArgumentSeedRandom) { 68 | list = append(list, "Random seed") 69 | } else { 70 | list = append(list, "Kmeans++") 71 | } 72 | // Mean or median 73 | if prominentcolor.IsBitSet(bits, prominentcolor.ArgumentAverageMean) { 74 | list = append(list, "Mean") 75 | } else { 76 | list = append(list, "Median") 77 | } 78 | // LAB or RGB 79 | if prominentcolor.IsBitSet(bits, prominentcolor.ArgumentLAB) { 80 | list = append(list, "LAB") 81 | } else { 82 | list = append(list, "RGB") 83 | } 84 | // Cropping or not 85 | if prominentcolor.IsBitSet(bits, prominentcolor.ArgumentNoCropping) { 86 | list = append(list, "No cropping") 87 | } else { 88 | list = append(list, "Cropping center") 89 | } 90 | // Done 91 | return strings.Join(list, ", ") 92 | } 93 | 94 | func main() { 95 | // Prepare 96 | outputDirectory := "./" 97 | dataDirectory := "./" 98 | 99 | var buff strings.Builder 100 | buff.WriteString("

Colors listed in order of dominance: hex color followed by number of entries

") 101 | 102 | // for each file within working directory 103 | files, err := ioutil.ReadDir(dataDirectory) 104 | if err != nil { 105 | log.Fatal(err) 106 | } 107 | for _, f := range files { 108 | filename := f.Name() 109 | // Only process jpg 110 | if !strings.HasSuffix(filename, ".jpg") { 111 | continue 112 | } 113 | // Define the differents sets of params 114 | kk := []int{ 115 | prominentcolor.ArgumentAverageMean | prominentcolor.ArgumentNoCropping, 116 | prominentcolor.ArgumentNoCropping, 117 | prominentcolor.ArgumentDefault, 118 | } 119 | // Load the image 120 | img, err := loadImage(filename) 121 | if err != nil { 122 | log.Printf("Error loading image %s\n", filename) 123 | log.Println(err) 124 | continue 125 | } 126 | // Process & html output 127 | buff.WriteString("") 130 | } 131 | 132 | // Finalize the html output 133 | buff.WriteString("
") 128 | buff.WriteString(processBatch(3, kk, img)) 129 | buff.WriteString("
") 134 | 135 | // And write it to the disk 136 | if err = ioutil.WriteFile(outputDirectory+"output.html", []byte(buff.String()), 0644); err != nil { 137 | panic(err) 138 | } 139 | } 140 | -------------------------------------------------------------------------------- /example/output.html: -------------------------------------------------------------------------------- 1 |

Colors listed in order of dominance: hex color followed by number of entries

K=3, Kmeans++, Mean, RGB, No cropping

#441F1C 1731#9E6B68 1512#CBB9B7 1077

K=3, Kmeans++, Median, RGB, No cropping

#401B18 1694#9B6B6A 1528#D2B2AE 1098

K=3, Kmeans++, Median, RGB, Cropping center

#6A1616 1701#E98584 1341#B64D50 1198

K=3, Kmeans++, Mean, RGB, No cropping

#676E73 5203#3A444A 2013#B1A98D 864

K=3, Kmeans++, Median, RGB, No cropping

#696F74 4839#3F4C50 2404#B4A78E 837

K=3, Kmeans++, Median, RGB, Cropping center

#414E52 3252#8A856F 2810#C0B293 2018

K=3, Kmeans++, Mean, RGB, No cropping

#364A2E 2213#7D7A64 1311#CEB6C0 796

K=3, Kmeans++, Median, RGB, No cropping

#364C31 2211#7D7860 1336#D6B9C8 773

K=3, Kmeans++, Median, RGB, Cropping center

#DAC8D1 1895#BF9695 752#6D5D4D 693

K=3, Kmeans++, Mean, RGB, No cropping

#5C8132 2038#A3B081 1254#242614 948

K=3, Kmeans++, Median, RGB, No cropping

#568531 1937#99B073 1373#1F2310 930

K=3, Kmeans++, Median, RGB, Cropping center

#607D35 2079#BBA38C 1234#2A2010 927

K=3, Kmeans++, Mean, RGB, No cropping

#39382D 1849#6B7280 1341#BABEAC 1050

K=3, Kmeans++, Median, RGB, No cropping

#373829 1862#727484 1407#BFC0B3 971

K=3, Kmeans++, Median, RGB, Cropping center

#6B7AB7 1674#1D222E 1328#364985 1238

K=3, Kmeans++, Mean, RGB, No cropping

#6E3F17 676#B57B4B 638#BEB0A1 295

K=3, Kmeans++, Median, RGB, No cropping

#C49565 595#673911 521#9B6738 493

K=3, Kmeans++, Median, RGB, Cropping center

#643811 1831#A16B3A 1218#CE9663 1126

K=3, Kmeans++, Mean, RGB, No cropping

#D0C018 300#DAD796 205#908F77 202

K=3, Kmeans++, Median, RGB, No cropping

#CFBE0C 290#979788 224#E2DB96 193

K=3, Kmeans++, Median, RGB, Cropping center

#EEEFE4 2611#D0C003 318#A0A18E 271

K=3, Kmeans++, Mean, RGB, No cropping

#474F57 2313#8361B0 2162#C1B236 1925

K=3, Kmeans++, Median, RGB, No cropping

#89649E 2492#43505A 2245#D1BD25 1663

K=3, Kmeans++, Median, RGB, Cropping center

#D9CC06 4351#BC88B8 1450#443027 599
-------------------------------------------------------------------------------- /example2/example2.go: -------------------------------------------------------------------------------- 1 | package main 2 | 3 | import ( 4 | "image" 5 | _ "image/jpeg" 6 | "log" 7 | "os" 8 | 9 | "path/filepath" 10 | 11 | "fmt" 12 | 13 | prominentcolor ".." 14 | ) 15 | 16 | func loadImage(fileInput string) (image.Image, error) { 17 | f, err := os.Open(fileInput) 18 | defer f.Close() 19 | if err != nil { 20 | log.Println("File not found:", fileInput) 21 | return nil, err 22 | } 23 | img, _, err := image.Decode(f) 24 | if err != nil { 25 | return nil, err 26 | } 27 | 28 | return img, nil 29 | } 30 | 31 | // Process images in a directory, for each image it picks out the dominant color and 32 | // prints out an imagemagick call to resize image and use the dominant color as padding for the background 33 | // it saves tmp files in /tmp/ with the masked bit marked as pink 34 | func main() { 35 | 36 | inputPattern := "../example/*.jpg" 37 | outputDirectory := "/tmp/" 38 | 39 | files, err := filepath.Glob(inputPattern) 40 | 41 | if nil != err { 42 | log.Println(err) 43 | log.Println("Error: failed glob") 44 | return 45 | } 46 | 47 | for _, file := range files { 48 | img, err := loadImage(file) 49 | if nil != err { 50 | log.Println(err) 51 | log.Printf("Error: failed loading %s\n", file) 52 | continue 53 | } 54 | cols, err := prominentcolor.KmeansWithArgs(prominentcolor.ArgumentNoCropping|prominentcolor.ArgumentDebugImage, img) 55 | if err != nil { 56 | log.Println(err) 57 | continue 58 | } 59 | col := cols[0].AsString() 60 | base := filepath.Base(file) 61 | fmt.Printf("convert %s -resize 800x356 -background '#%s' -gravity center -extent 800x356 %s%s\n", base, col, outputDirectory, base) 62 | } 63 | 64 | } 65 | -------------------------------------------------------------------------------- /imgprep.go: -------------------------------------------------------------------------------- 1 | // Copyright 2016 Carl Asman. All rights reserved. 2 | // Use of this source code is governed by a BSD-style 3 | // license that can be found in the LICENSE file. 4 | 5 | package prominentcolor 6 | 7 | import ( 8 | "image" 9 | "image/color" 10 | "image/draw" 11 | "log" 12 | 13 | "image/jpeg" 14 | "os" 15 | 16 | "time" 17 | 18 | "fmt" 19 | 20 | "github.com/nfnt/resize" 21 | "github.com/oliamb/cutter" 22 | ) 23 | 24 | // ColorBackgroundMask defines which color channels to look for color to ignore 25 | type ColorBackgroundMask struct { 26 | // Setting them all to true or all to false; Treshold is used, otherwise PercDiff 27 | R, G, B bool 28 | 29 | // Treshold is the lower limit to check against for each r,g,b value, when all R,G,B that has true set should be above to be ignored (upper if all set to false) 30 | Treshold uint32 31 | 32 | // PercDiff if any of R,G,B is true (but not all), any of the other colors divided by the color value that is true, must be below PercDiff 33 | PercDiff float32 34 | } 35 | 36 | // ProcessImg process the image and mark unwanted pixels transparent. 37 | // It checks the corners, if not all of them match the mask, we conclude it's not a clipart/solid background and do nothing 38 | func ProcessImg(arguments int, bgmasks []ColorBackgroundMask, img image.Image) draw.Image { 39 | imgDraw := createDrawImage(img) 40 | rect := imgDraw.Bounds() 41 | 42 | //loop through the masks, and the first one that matches on the four corners is the one that will be used 43 | foundMaskThatmatched := false 44 | var bgmaskToUse ColorBackgroundMask 45 | for _, bgmask := range bgmasks { 46 | // Check the corners, if not all of them are the color of the mask, 47 | // we conclude it's not a solid background and do nothing special 48 | if !ignorePixel(rect.Min.X, rect.Min.Y, bgmask, &imgDraw) || !ignorePixel(rect.Min.X, rect.Max.Y-1, bgmask, &imgDraw) || !ignorePixel(rect.Max.X-1, rect.Min.Y, bgmask, &imgDraw) || !ignorePixel(rect.Max.X-1, rect.Max.Y-1, bgmask, &imgDraw) { 49 | continue 50 | } 51 | foundMaskThatmatched = true 52 | bgmaskToUse = bgmask 53 | } 54 | 55 | // no mask that we can apply 56 | if !foundMaskThatmatched { 57 | return imgDraw 58 | } 59 | 60 | ProcessImgOutline(bgmaskToUse, &imgDraw) 61 | 62 | // if debug argument is set, save a tmp file to be able to view what was masked out 63 | if IsBitSet(arguments, ArgumentDebugImage) { 64 | tmpFilename := fmt.Sprintf("/tmp/tmp%d.jpg", time.Now().UnixNano()/1000000) 65 | toimg, _ := os.Create(tmpFilename) 66 | defer toimg.Close() 67 | jpeg.Encode(toimg, imgDraw, &jpeg.Options{Quality: 100}) 68 | } 69 | 70 | return imgDraw 71 | } 72 | 73 | // ProcessImgOutline follow the outline of the image and mark all "white" pixels as transparent 74 | func ProcessImgOutline(bgmask ColorBackgroundMask, imgDraw *draw.Image) { 75 | 76 | rect := (*imgDraw).Bounds() 77 | 78 | var pointsToProcess []image.Point 79 | 80 | // points to add to start processing: corners only 81 | pointsToProcess = append(pointsToProcess, image.Point{X: rect.Min.X, Y: rect.Min.Y}) 82 | pointsToProcess = append(pointsToProcess, image.Point{X: rect.Min.X, Y: rect.Max.Y - 1}) 83 | pointsToProcess = append(pointsToProcess, image.Point{X: rect.Max.X - 1, Y: rect.Min.Y}) 84 | pointsToProcess = append(pointsToProcess, image.Point{X: rect.Max.X - 1, Y: rect.Max.Y - 1}) 85 | 86 | var p image.Point 87 | for len(pointsToProcess) > 0 { 88 | //pop from slice 89 | p, pointsToProcess = pointsToProcess[len(pointsToProcess)-1], pointsToProcess[:len(pointsToProcess)-1] 90 | 91 | if !isPixelTransparent(p.X, p.Y, imgDraw) && ignorePixel(p.X, p.Y, bgmask, (imgDraw)) { 92 | 93 | //Mark the pixel 94 | markPixel(p.X, p.Y, (imgDraw)) 95 | if !isPixelTransparent(p.X, p.Y, imgDraw) { 96 | log.Println("ERROR: marking") 97 | } 98 | 99 | //add pixels above, below, left,right 100 | //unless its transparent 101 | if rect.Min.X < p.X { 102 | if !isPixelTransparent(p.X-1, p.Y, imgDraw) { 103 | pointsToProcess = append(pointsToProcess, image.Point{X: p.X - 1, Y: p.Y}) 104 | } 105 | } 106 | 107 | if p.X < rect.Max.X-1 { 108 | if !isPixelTransparent(p.X+1, p.Y, imgDraw) { 109 | pointsToProcess = append(pointsToProcess, image.Point{X: p.X + 1, Y: p.Y}) 110 | } 111 | } 112 | 113 | if rect.Min.Y < p.Y { 114 | if !isPixelTransparent(p.X, p.Y-1, imgDraw) { 115 | pointsToProcess = append(pointsToProcess, image.Point{X: p.X, Y: p.Y - 1}) 116 | } 117 | } 118 | 119 | if p.Y < rect.Max.Y-1 { 120 | if !isPixelTransparent(p.X, p.Y+1, imgDraw) { 121 | pointsToProcess = append(pointsToProcess, image.Point{X: p.X, Y: p.Y + 1}) 122 | } 123 | } 124 | } 125 | } 126 | } 127 | 128 | // createDrawImage creates a draw.Image so we can work with the single pixels 129 | func createDrawImage(img image.Image) draw.Image { 130 | b := img.Bounds() 131 | cimg := image.NewRGBA(b) 132 | draw.Draw(cimg, b, img, b.Min, draw.Src) 133 | return cimg 134 | } 135 | 136 | // prepareImg resizes to a smaller size and remove any "white" background pixels for isolated/clipart images 137 | func prepareImg(arguments int, bgmasks []ColorBackgroundMask, imageSize uint, orgimg image.Image) image.Image { 138 | 139 | if !IsBitSet(arguments, ArgumentNoCropping) { 140 | // crop to remove 25% on all sides 141 | croppedimg, err := cutter.Crop(orgimg, cutter.Config{ 142 | Width: int(orgimg.Bounds().Dx() / 2), 143 | Height: int(orgimg.Bounds().Dy() / 2), 144 | Mode: cutter.Centered, 145 | }) 146 | 147 | if err != nil { 148 | log.Println("Warning: failed cropping") 149 | log.Println(err) 150 | } else { 151 | orgimg = croppedimg 152 | } 153 | } 154 | 155 | // Don't resize if the image is smaller than imageSize 156 | rec := orgimg.Bounds() 157 | 158 | if uint(rec.Dx()) > imageSize || uint(rec.Dy()) > imageSize { 159 | img := resize.Resize(imageSize, 0, orgimg, resize.Lanczos3) 160 | return ProcessImg(arguments, bgmasks, img) 161 | } 162 | 163 | return ProcessImg(arguments, bgmasks, orgimg) 164 | } 165 | 166 | // markPixel sets a purple color (to make it stick out if we want to look at the image) and makes the pixel transparent 167 | func markPixel(x, y int, img *draw.Image) { 168 | (*img).Set(x, y, color.RGBA{255, 0, 255, 0}) 169 | } 170 | 171 | // isPixelTransparent returns bool if the pixel is transparent (alpha==0) 172 | func isPixelTransparent(x, y int, img *draw.Image) bool { 173 | colorAt := (*img).At(x, y) 174 | _, _, _, a := colorAt.RGBA() 175 | return a == 0 176 | } 177 | 178 | // ignorePixel checks if the pixel should be ignored (i.e. being transparent or white) 179 | func ignorePixel(x, y int, bgmask ColorBackgroundMask, img *draw.Image) bool { 180 | colorAt := (*img).At(x, y) 181 | 182 | r, g, b, a := colorAt.RGBA() 183 | 184 | if a == 0 { 185 | return true 186 | } 187 | 188 | //if looking for black 189 | if !(bgmask.R || bgmask.G || bgmask.B) { 190 | if r > bgmask.Treshold { 191 | return false 192 | } 193 | 194 | if g > bgmask.Treshold { 195 | return false 196 | } 197 | 198 | if b > bgmask.Treshold { 199 | return false 200 | } 201 | 202 | return true 203 | } 204 | 205 | //if not looking for white 206 | if !(bgmask.R && bgmask.G && bgmask.B) { 207 | 208 | var aArr, baseArr []float32 209 | 210 | if bgmask.R { 211 | baseArr = append(baseArr, float32(r)) 212 | } else { 213 | aArr = append(aArr, float32(r)) 214 | } 215 | if bgmask.G { 216 | baseArr = append(baseArr, float32(g)) 217 | } else { 218 | aArr = append(aArr, float32(g)) 219 | } 220 | if bgmask.B { 221 | baseArr = append(baseArr, float32(b)) 222 | } else { 223 | aArr = append(aArr, float32(b)) 224 | } 225 | 226 | for _, val := range aArr { 227 | for _, base := range baseArr { 228 | if val/base > bgmask.PercDiff { 229 | return false 230 | } 231 | } 232 | } 233 | 234 | return true 235 | } 236 | 237 | // Checking for white 238 | 239 | if bgmask.R && r < bgmask.Treshold { 240 | return false 241 | } 242 | 243 | if bgmask.G && g < bgmask.Treshold { 244 | return false 245 | } 246 | 247 | if bgmask.B && b < bgmask.Treshold { 248 | return false 249 | } 250 | 251 | return true 252 | } 253 | -------------------------------------------------------------------------------- /kmeans.go: -------------------------------------------------------------------------------- 1 | // Copyright 2016 Carl Asman. All rights reserved. 2 | // Use of this source code is governed by a BSD-style 3 | // license that can be found in the LICENSE file. 4 | 5 | // Package prominentcolor finds the K most dominant/prominent colors in an image 6 | package prominentcolor 7 | 8 | import ( 9 | "fmt" 10 | "image" 11 | "image/color" 12 | "log" 13 | "math/rand" 14 | 15 | "sort" 16 | 17 | "time" 18 | 19 | "github.com/lucasb-eyer/go-colorful" 20 | ) 21 | 22 | const ( 23 | // ArgumentDefault default settings 24 | ArgumentDefault int = 0 25 | // ArgumentSeedRandom randomly pick initial values (instead of K-means++) 26 | ArgumentSeedRandom = 1 << iota 27 | // ArgumentAverageMean take the mean value when determining the centroid color (instead of median) 28 | ArgumentAverageMean 29 | // ArgumentNoCropping do not crop background that is considered "white" 30 | ArgumentNoCropping 31 | // ArgumentLAB (experimental, it seems to be buggy in some cases): uses LAB instead of RGB when measuring distance 32 | ArgumentLAB 33 | // ArgumentDebugImage saves a tmp file in /tmp/ where the area that has been cut away by the mask is marked pink 34 | // useful when figuring out what values to pick for the masks 35 | ArgumentDebugImage 36 | ) 37 | 38 | const ( 39 | // DefaultK is the k used as default 40 | DefaultK = 3 41 | // DefaultSize is the default size images are re-sized to 42 | DefaultSize = 80 43 | ) 44 | 45 | var ( 46 | // MaskWhite "constant" for white mask (for ease of re-use for other mask arrays) 47 | MaskWhite = ColorBackgroundMask{R: true, G: true, B: true, Treshold: uint32(0xc000)} 48 | // MaskBlack "constant" for black mask (for ease of re-use for other mask arrays) 49 | MaskBlack = ColorBackgroundMask{R: false, G: false, B: false, Treshold: uint32(0x5000)} 50 | // MaskGreen "constant" for green mask (for ease of re-use for other mask arrays) 51 | MaskGreen = ColorBackgroundMask{R: false, G: true, B: false, PercDiff: 0.9} 52 | ) 53 | 54 | // ErrNoPixelsFound is returned when no non-alpha pixels are found in the provided image 55 | var ErrNoPixelsFound = fmt.Errorf("Failed, no non-alpha pixels found (either fully transparent image, or the ColorBackgroundMask removed all pixels)") 56 | 57 | // ColorRGB contains the color values 58 | type ColorRGB struct { 59 | R, G, B uint32 60 | } 61 | 62 | // ColorItem contains color and have many occurrences of this color found 63 | type ColorItem struct { 64 | Color ColorRGB 65 | Cnt int 66 | } 67 | 68 | // AsString gives back the color in hex as 6 character string 69 | func (c *ColorItem) AsString() string { 70 | return fmt.Sprintf("%.2X%.2X%.2X", c.Color.R, c.Color.G, c.Color.B) 71 | } 72 | 73 | // createColor returns ColorItem struct unless it was a transparent color 74 | func createColor(c color.Color) (ColorItem, bool) { 75 | r, g, b, a := c.RGBA() 76 | 77 | if a == 0 { 78 | // transparent pixels are ignored 79 | return ColorItem{}, true 80 | } 81 | 82 | divby := uint32(256.0) 83 | return ColorItem{Color: ColorRGB{R: r / divby, G: g / divby, B: b / divby}}, false 84 | } 85 | 86 | // IsBitSet check if "lookingfor" is set in "bitset" 87 | func IsBitSet(bitset int, lookingfor int) bool { 88 | return lookingfor == (bitset & lookingfor) 89 | } 90 | 91 | // GetDefaultMasks returns the masks that are used for the default settings 92 | func GetDefaultMasks() []ColorBackgroundMask { 93 | return []ColorBackgroundMask{MaskWhite, MaskBlack, MaskGreen} 94 | } 95 | 96 | // Kmeans uses the default: k=3, Kmeans++, Median, crop center, resize to 80 pixels, mask out white/black/green backgrounds 97 | // It returns an array of ColorItem which are three centroids, sorted according to dominance (most frequent first). 98 | func Kmeans(orgimg image.Image) (centroids []ColorItem, err error) { 99 | return KmeansWithAll(DefaultK, orgimg, ArgumentDefault, DefaultSize, GetDefaultMasks()) 100 | } 101 | 102 | // KmeansWithArgs takes arguments which consists of the bits, see constants Argument* 103 | func KmeansWithArgs(arguments int, orgimg image.Image) (centroids []ColorItem, err error) { 104 | return KmeansWithAll(DefaultK, orgimg, arguments, DefaultSize, GetDefaultMasks()) 105 | } 106 | 107 | // KmeansWithAll takes additional arguments to define k, arguments (see constants Argument*), size to resize and masks to use 108 | func KmeansWithAll(k int, orgimg image.Image, arguments int, imageReSize uint, bgmasks []ColorBackgroundMask) ([]ColorItem, error) { 109 | 110 | img := prepareImg(arguments, bgmasks, imageReSize, orgimg) 111 | 112 | allColors, _ := extractColorsAsArray(img) 113 | 114 | numColors := len(allColors) 115 | 116 | if numColors == 0 { 117 | return nil, ErrNoPixelsFound 118 | } 119 | 120 | if numColors == 1 { 121 | return allColors, nil 122 | } 123 | 124 | if numColors <= k { 125 | sortCentroids(allColors) 126 | return allColors, nil 127 | } 128 | 129 | centroids, err := kmeansSeed(k, allColors, arguments) 130 | if err != nil { 131 | return nil, err 132 | } 133 | 134 | cent := make([][]ColorItem, k) 135 | 136 | //initialize 137 | cent[0] = allColors 138 | for i := 1; i < k; i++ { 139 | cent[i] = []ColorItem{} 140 | } 141 | 142 | //rounds is a safety net to make sure we terminate if its a bug in our distance function (or elsewhere) that makes k-means not terminate 143 | rounds := 0 144 | maxRounds := 5000 145 | changes := 1 146 | 147 | for changes > 0 && rounds < maxRounds { 148 | changes = 0 149 | tmpCent := make([][]ColorItem, k) 150 | for i := 0; i < k; i++ { 151 | tmpCent[i] = []ColorItem{} 152 | } 153 | 154 | for i := 0; i < k; i++ { 155 | for _, aColor := range cent[i] { 156 | closestCentroid := findClosest(arguments, aColor, centroids) 157 | 158 | tmpCent[closestCentroid] = append(tmpCent[closestCentroid], aColor) 159 | if closestCentroid != i { 160 | changes++ 161 | } 162 | } 163 | } 164 | cent = tmpCent 165 | centroids = calculateCentroids(cent, arguments) 166 | rounds++ 167 | } 168 | 169 | if rounds >= maxRounds { 170 | log.Println("Warning: terminated k-means due to max number of iterations") 171 | } 172 | 173 | sortCentroids(centroids) 174 | return centroids, nil 175 | } 176 | 177 | // ByColorCnt makes the ColorItem sortable 178 | type byColorCnt []ColorItem 179 | 180 | func (a byColorCnt) Len() int { return len(a) } 181 | func (a byColorCnt) Swap(i, j int) { a[i], a[j] = a[j], a[i] } 182 | func (a byColorCnt) Less(i, j int) bool { 183 | if a[i].Cnt == a[j].Cnt { 184 | return a[i].AsString() < a[j].AsString() 185 | } 186 | return a[i].Cnt < a[j].Cnt 187 | } 188 | 189 | // sortCentroids sorts them from most dominant color descending 190 | func sortCentroids(centroids []ColorItem) { 191 | sort.Sort(sort.Reverse(byColorCnt(centroids))) 192 | } 193 | 194 | func calculateCentroids(cent [][]ColorItem, arguments int) []ColorItem { 195 | var centroids []ColorItem 196 | 197 | for _, colors := range cent { 198 | 199 | var meanColor ColorItem 200 | if IsBitSet(arguments, ArgumentAverageMean) { 201 | meanColor = mean(colors) 202 | } else { 203 | meanColor = median(colors) 204 | } 205 | 206 | centroids = append(centroids, meanColor) 207 | } 208 | 209 | return centroids 210 | } 211 | 212 | // mean calculate the mean color values from an array of colors 213 | func mean(colors []ColorItem) ColorItem { 214 | 215 | var r, g, b float64 216 | 217 | r, g, b = 0.0, 0.0, 0.0 218 | 219 | cntInThisBucket := 0 220 | for _, aColor := range colors { 221 | cntInThisBucket += aColor.Cnt 222 | r += float64(aColor.Color.R) 223 | g += float64(aColor.Color.G) 224 | b += float64(aColor.Color.B) 225 | } 226 | 227 | theSize := float64(len(colors)) 228 | 229 | return ColorItem{Cnt: cntInThisBucket, Color: ColorRGB{R: uint32(r / theSize), G: uint32(g / theSize), B: uint32(b / theSize)}} 230 | } 231 | 232 | // median calculate the median color from an array of colors 233 | func median(colors []ColorItem) ColorItem { 234 | 235 | var rValues, gValues, bValues []int 236 | 237 | cntInThisBucket := 0 238 | 239 | for _, aColor := range colors { 240 | cntInThisBucket += aColor.Cnt 241 | rValues = append(rValues, int(aColor.Color.R)) 242 | gValues = append(gValues, int(aColor.Color.G)) 243 | bValues = append(bValues, int(aColor.Color.B)) 244 | } 245 | 246 | retR := 0 247 | if 0 != len(rValues) { 248 | sort.Ints(rValues) 249 | retR = rValues[int(len(rValues)/2)] 250 | } 251 | 252 | retG := 0 253 | if 0 != len(gValues) { 254 | sort.Ints(gValues) 255 | retG = gValues[int(len(gValues)/2)] 256 | } 257 | 258 | retB := 0 259 | if 0 != len(bValues) { 260 | sort.Ints(bValues) 261 | retB = bValues[int(len(bValues)/2)] 262 | } 263 | 264 | return ColorItem{Cnt: cntInThisBucket, Color: ColorRGB{R: uint32(retR), G: uint32(retG), B: uint32(retB)}} 265 | } 266 | 267 | // extractColorsAsArray counts the number of occurrences of each color in the image, returns array and numPixels 268 | func extractColorsAsArray(img image.Image) ([]ColorItem, int) { 269 | m, numPixels := extractColors(img) 270 | v := make([]ColorItem, len(m)) 271 | idx := 0 272 | for _, value := range m { 273 | v[idx] = value 274 | idx++ 275 | } 276 | 277 | return v, numPixels 278 | } 279 | 280 | // extractColors counts the number of occurrences of each color in the image, returns map 281 | func extractColors(img image.Image) (map[string]ColorItem, int) { 282 | 283 | m := make(map[string]ColorItem) 284 | 285 | numPixels := 0 286 | data := img.Bounds() 287 | for x := data.Min.X; x < data.Max.X; x++ { 288 | for y := data.Min.Y; y < data.Max.Y; y++ { 289 | colorAt := img.At(x, y) 290 | colorItem, ignore := createColor(colorAt) 291 | if ignore { 292 | continue 293 | } 294 | numPixels++ 295 | asString := colorItem.AsString() 296 | value, ok := m[asString] 297 | if ok { 298 | value.Cnt++ 299 | m[asString] = value 300 | } else { 301 | colorItem.Cnt = 1 302 | m[asString] = colorItem 303 | } 304 | } 305 | } 306 | return m, numPixels 307 | } 308 | 309 | // findClosest returns the index of the closest centroid to the color "c" 310 | func findClosest(arguments int, c ColorItem, centroids []ColorItem) int { 311 | 312 | centLen := len(centroids) 313 | 314 | closestIdx := 0 315 | closestDistance := distance(arguments, c, centroids[0]) 316 | 317 | for i := 1; i < centLen; i++ { 318 | distance := distance(arguments, c, centroids[i]) 319 | if distance < closestDistance { 320 | closestIdx = i 321 | closestDistance = distance 322 | } 323 | } 324 | return closestIdx 325 | } 326 | 327 | // distance returns the distance between two colors 328 | func distance(arguments int, c ColorItem, p ColorItem) float64 { 329 | if IsBitSet(arguments, ArgumentLAB) { 330 | return distanceLAB(c, p) 331 | } 332 | return distanceRGB(c, p) 333 | } 334 | 335 | func distanceLAB(c ColorItem, p ColorItem) float64 { 336 | errmsg := "Warning: LAB failed, fallback to RGB" 337 | 338 | a, err := colorful.Hex("#" + c.AsString()) 339 | if err != nil { 340 | log.Fatal(err) 341 | log.Println(errmsg) 342 | return distanceRGB(c, p) 343 | } 344 | 345 | b, err2 := colorful.Hex("#" + p.AsString()) 346 | if err2 != nil { 347 | log.Fatal(err2) 348 | log.Println(errmsg) 349 | return distanceRGB(c, p) 350 | } 351 | 352 | return a.DistanceLab(b) 353 | } 354 | 355 | func distanceRGB(c ColorItem, p ColorItem) float64 { 356 | r := c.Color.R 357 | g := c.Color.G 358 | b := c.Color.B 359 | 360 | r2 := p.Color.R 361 | g2 := p.Color.G 362 | b2 := p.Color.B 363 | 364 | //sqrt not needed since we just want to compare distances to each other 365 | return float64((r-r2)*(r-r2) + (g-g2)*(g-g2) + (b-b2)*(b-b2)) 366 | } 367 | 368 | // kmeansSeed calculates the initial cluster centroids 369 | func kmeansSeed(k int, allColors []ColorItem, arguments int) ([]ColorItem, error) { 370 | if k > len(allColors) { 371 | return nil, fmt.Errorf("Failed, k larger than len(allColors): %d vs %d\n", k, len(allColors)) 372 | } 373 | 374 | rand.Seed(time.Now().UnixNano()) 375 | 376 | if IsBitSet(arguments, ArgumentSeedRandom) { 377 | return kmeansSeedRandom(k, allColors), nil 378 | } 379 | return kmeansPlusPlusSeed(k, arguments, allColors), nil 380 | } 381 | 382 | // kmeansSeedRandom picks k random points as initial centroids 383 | func kmeansSeedRandom(k int, allColors []ColorItem) []ColorItem { 384 | var centroids []ColorItem 385 | 386 | taken := make(map[int]bool) 387 | 388 | for i := 0; i < k; i++ { 389 | idx := rand.Intn(len(allColors)) 390 | 391 | //check if we already taken this one 392 | _, ok := taken[idx] 393 | if ok { 394 | i-- 395 | continue 396 | } 397 | taken[idx] = true 398 | centroids = append(centroids, allColors[idx]) 399 | } 400 | return centroids 401 | } 402 | 403 | // kmeansPlusPlusSeed picks initial centroids using K-Means++ 404 | func kmeansPlusPlusSeed(k int, arguments int, allColors []ColorItem) []ColorItem { 405 | var centroids []ColorItem 406 | 407 | taken := make(map[int]bool) 408 | 409 | initIdx := rand.Intn(len(allColors)) 410 | centroids = append(centroids, allColors[initIdx]) 411 | taken[initIdx] = true 412 | 413 | for kk := 1; kk < k; kk++ { 414 | 415 | totaldistances := 0.0 416 | var point2distance []float64 417 | 418 | for j := 0; j < len(allColors); j++ { 419 | 420 | _, ok := taken[j] 421 | if ok { 422 | point2distance = append(point2distance, 0.0) 423 | continue 424 | } 425 | 426 | minDistanceToCluster := -1.0 427 | for i := 0; i < len(centroids); i++ { 428 | d := distance(arguments, centroids[i], allColors[j]) 429 | if minDistanceToCluster == -1.0 || d < minDistanceToCluster { 430 | minDistanceToCluster = d 431 | } 432 | } 433 | 434 | squareDistance := minDistanceToCluster * minDistanceToCluster 435 | totaldistances += squareDistance 436 | point2distance = append(point2distance, squareDistance) 437 | } 438 | 439 | rndpoint := rand.Float64() * totaldistances 440 | 441 | sofar := 0.0 442 | for j := 0; j < len(point2distance); j++ { 443 | if rndpoint <= sofar { 444 | centroids = append(centroids, allColors[j]) 445 | taken[j] = true 446 | break 447 | } 448 | sofar += point2distance[j] 449 | } 450 | } 451 | 452 | return centroids 453 | } 454 | --------------------------------------------------------------------------------