├── .gitignore ├── Dockerfile ├── README.md ├── answer_card.jpg ├── app.py ├── detection_choice_question.py ├── detection_exam_num.py ├── main.py ├── notebook ├── answer_card_detection.ipynb └── choice_option_detection.ipynb ├── pic ├── answer_card.jpg ├── choice_question_detection_1.jpg ├── choice_question_detection_2.png ├── choice_question_detection_3.png ├── choice_question_detection_4.png ├── choice_question_detection_5.png ├── num_card_detection_1.png ├── num_card_detection_2.png ├── num_card_detection_3.png ├── special_choice_question.png ├── special_num_card_1.png ├── special_num_card_2.png ├── sub_answer_card_1.jpg ├── sub_answer_card_2.jpg ├── sub_answer_card_3.jpg ├── sub_answer_card_4.jpg ├── sub_answer_card_5.jpg ├── sub_answer_card_6.jpg ├── sub_answer_card_7.jpg └── sub_answer_card_8.jpg ├── requirements.txt ├── settings.py ├── test ├── test_cnocr.py └── test_pytesseract.py └── utils.py /.gitignore: -------------------------------------------------------------------------------- 1 | *.jpg 2 | *.png 3 | *.pyc 4 | *.vscode 5 | *.ipynb_checkpoints 6 | *.idea -------------------------------------------------------------------------------- /Dockerfile: -------------------------------------------------------------------------------- 1 | FROM registry-vpc.cn-shenzhen.aliyuncs.com/spark-base/anaconda3:5.2.0 2 | 3 | WORKDIR /app 4 | 5 | COPY requirements.txt requirements.txt 6 | 7 | RUN set -x \ 8 | && pip install --no-cache-dir -r requirements.txt 9 | 10 | COPY . . 11 | 12 | CMD ["tail","-f","/dev/stderr"] 13 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # teaching-answer-card-tool 2 | 3 | 答题卡检测 4 | 5 | ### 1、多张图连续检测 6 | 功能：如果一套试卷的答题卡数量大于1，也支持检测，并且按试题顺序拼凑检测结果（即下面这套试卷假设存在位于第二张答题卡上的第八题，支持连续检测） 7 | 8 | ![](pic/answer_card.jpg) 9 | 10 | ### 2、试题切割 11 | 功能： 12 | - 将答题卡按试题切成一张张图片，并将图片与题序一一对应 13 | - 程序能知道第六大题存在2张图片 14 | 15 | 程序执行结果： 16 | 第一题 17 | ![](pic/sub_answer_card_1.jpg) 18 | 19 | 第二题 20 | ![](pic/sub_answer_card_2.jpg) 21 | 22 | 第三题 23 | ![](pic/sub_answer_card_3.jpg) 24 | 25 | 第四题 26 | ![](pic/sub_answer_card_4.jpg) 27 | 28 | 第五题 29 | ![](pic/sub_answer_card_5.jpg) 30 | 31 | 第六题（程序能知道第六大题存在2张图片） 32 | ![](pic/sub_answer_card_6.jpg) 33 | 34 | ![](pic/sub_answer_card_7.jpg) 35 | 36 | 第七题： 37 | 38 | ![](pic/sub_answer_card_8.jpg) 39 | 40 | 41 | ### 3、学生考号识别 42 | #### 1）切割出学生手写的准考证号 43 | ![](pic/num_card_detection_1.png) 44 | 45 | #### 2）识别出学生填充的准考证号 46 | 47 | 识别结果：53311 48 | 49 | ![](pic/num_card_detection_2.png) 50 | 51 | ![](pic/num_card_detection_3.png) 52 | 53 | 54 | ### 4、选择题自动批改检测 55 | 识别结果（key：题序，value：答案的索引）： 56 | 57 | {1: [2], 2: [1], 3: [3], 4: [3], 5: [1], 6: [1], 7: [4], 8: [2], 9: [4], 10: [1], 11: [3], 12: [1], 13: [1], 14: [1], 15: [1]} 58 | 59 | 选择题自动批改支持以下变化 60 | - 无论试题的候选项有四个还是五六七八个，都支持识别 61 | - 答案无论是一个还是多个，都支持识别 62 | - 选择题，无论是竖着排列，还是横着排列，还是乱序排序，支持识别 63 | 64 | 65 | 原题： 66 | ![](pic/choice_question_detection_1.jpg) 67 | 68 | 识别出所有的轮廓： 69 | 70 | ![](pic/choice_question_detection_2.png) 71 | 72 | 筛选出答案： 73 | 74 | ![](pic/choice_question_detection_3.png) 75 | 76 | 筛选出候选框： 77 | 78 | ![](pic/choice_question_detection_4.png) 79 | 80 | 筛选出题序： 81 | 82 | ![](pic/choice_question_detection_5.png) 83 | 84 | 根据题序及其坐标，寻找到对应题序的答案与候选框，进行自动批改 85 | 86 | 87 | ### 5、特殊逻辑处理 88 | （1）填充区域出现连接的情况，只要连接的不是太过分，都支持识别 89 | 90 | ![](pic/special_num_card_1.png) 91 | 92 | ![](pic/special_num_card_2.png) 93 | 94 | （2）填充区域，先填充，再擦除，存在曾经填充过的痕迹，目前已经优化，支持准确识别 95 | 96 | ![](pic/special_choice_question.png) 97 | 98 | 99 | -------------------------------------------------------------------------------- /answer_card.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Wasim37/answer_card_detection/b45b2c4fc06f757a78eb0b16188b939bdb17267f/answer_card.jpg -------------------------------------------------------------------------------- /app.py: -------------------------------------------------------------------------------- 1 | ''' 2 | Author: wangxin 3 | Date: 2021-05-25 12:12:52 4 | LastEditTime: 2021-07-01 14:00:58 5 | LastEditors: Please set LastEditors 6 | Description: In User Settings Edit 7 | ''' 8 | 9 | from flask import Flask 10 | 11 | 12 | app = Flask(__name__) 13 | 14 | 15 | @app.route('/detection', methods=["POST"]) 16 | def answer_card_detection(): 17 | return "我是一个接口" 18 | 19 | 20 | # python3 -m flask run 21 | if __name__ == '__main__': 22 | app.run(host='0.0.0.0', port=8890, debug=True) 23 | -------------------------------------------------------------------------------- /detection_choice_question.py: -------------------------------------------------------------------------------- 1 | ''' 2 | Author: wangxin 3 | Date: 2021-05-25 10:31:01 4 | LastEditTime: 2021-07-01 14:11:58 5 | LastEditors: Please set LastEditors 6 | Description: 选择题自动识别与批改 7 | ''' 8 | 9 | # coding=utf-8 10 | import cv2 11 | import numpy as np 12 | import pytesseract 13 | from pytesseract import Output 14 | from settings import ANSWER_CARD_MIN_WIDTH, ANSWER_CARD_SIZE 15 | from utils import sort_contours, save_img_by_cnts 16 | 17 | 18 | def detection_choice_question(images_path, ocr): 19 | """ 选择题自动识别与批改 20 | 21 | Args: 22 | images_path (list): 图片地址列表 23 | Returns: 24 | [list]: 每张图片的识别结果 25 | """ 26 | 27 | sub_answer_cnt_szie = 0 28 | question_answers = [] 29 | for img_path in images_path: 30 | image = cv2.imread(img_path) 31 | if not is_choice_question(image): 32 | continue 33 | 34 | # 获取图片中填充的全部答案轮廓 35 | answer_option_cnts = get_answer_option_cnts(image) 36 | if len(answer_option_cnts) > 0: 37 | save_img_by_cnts('out/answer_cnt_' + str(sub_answer_cnt_szie) + '.png', image.shape[:2], answer_option_cnts) 38 | 39 | # 所有被填充的选择项的中心的x坐标 40 | answer_options_center_x = get_cnt_center_x(answer_option_cnts) 41 | # 所有未被填充的选择项的中心的x坐标 42 | choice_options_center_x = get_choice_option_center_x(img_path) 43 | # 所有选择项的中心的x坐标 44 | all_options_center_x = answer_options_center_x + choice_options_center_x 45 | 46 | # 获取所有选择项的轮廓及其题序轮廓 47 | all_choice_option_cnts, question_number_cnts = get_choice_option_cnts(image, all_options_center_x) 48 | if len(all_choice_option_cnts) > 0: 49 | save_img_by_cnts('out/choice_cnt_' + str(sub_answer_cnt_szie) + '.png', image.shape[:2], all_choice_option_cnts) 50 | save_img_by_cnts('out/ques_num_' + str(sub_answer_cnt_szie) + '.png', image.shape[:2], question_number_cnts) 51 | 52 | sub_answer_cnt_szie = sub_answer_cnt_szie + 1 53 | 54 | # 选择题自动批改 55 | if len(all_choice_option_cnts) > 0: 56 | question_answer_dict = get_choice_question_answer_index(image, all_choice_option_cnts, answer_option_cnts, question_number_cnts, ocr) 57 | question_answers.append(question_answer_dict) 58 | return question_answers 59 | 60 | 61 | def get_answer_option_cnts(img): 62 | """ 识别图片中的填充的全部答案轮廓 63 | 64 | Args: 65 | img_path (String): 图片 66 | 67 | Returns: 68 | [list]: 候选项轮廓 69 | """ 70 | 71 | # 转灰度 72 | gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) 73 | # OTSU二值化（黑底白字） 74 | thresh_img = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)[1] 75 | 76 | # 腐蚀 77 | kernel = np.ones((5, 5), np.uint8) 78 | erode_img = cv2.erode(thresh_img, kernel, iterations=1) 79 | # 膨胀 80 | kernel = np.ones((9, 9), np.uint8) 81 | dilate_img = cv2.dilate(erode_img, kernel, iterations=1) 82 | 83 | # 提取答案的轮廓 84 | answer_cnts, _ = cv2.findContours(dilate_img.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE) 85 | 86 | # 减少答案轮廓的边数 87 | answer_option_cnts = [] 88 | for cnt in answer_cnts: 89 | peri = cv2.arcLength(cnt, True) 90 | approx = cv2.approxPolyDP(cnt, 0.06 * peri, True) 91 | answer_option_cnts.append(approx) 92 | 93 | # self.assertTrue(choiceAnswerCnts % 4 == 0, "候选框提取异常, 提取的数量不是4的整数") 94 | return answer_option_cnts 95 | 96 | 97 | def get_choice_option_cnts(img, all_options_center_x): 98 | """识别图片中的所有的选择项轮廓与题序轮廓 99 | 100 | Args: 101 | img ([type]): [description] 102 | all_option_center_x ([type]): [description] 103 | 104 | Returns: 105 | [type]: [description] 106 | """ 107 | # 灰度 108 | gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) 109 | 110 | # 二值化（黑底白字） 111 | thresh_img = cv2.threshold(gray, 0, 255, cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)[1] 112 | 113 | # 对高亮部分膨胀 114 | # 因为候选区域由三部分组成（左括号、右括号、大写的英文字母），通过膨胀将三个区域连成一片 115 | kernel = np.ones((11, 11), np.uint8) 116 | dilate_img = cv2.dilate(thresh_img, kernel, iterations=1) 117 | 118 | # 提取膨胀后的轮廓 119 | option_cnts, _ = cv2.findContours(dilate_img.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE) 120 | 121 | # 所有候选框的轮廓 122 | choice_option_cnts = [] 123 | # 每道选择题的题序 124 | question_number_cnts = [] 125 | for c in option_cnts: 126 | peri = cv2.arcLength(c, True) 127 | area = cv2.contourArea(c) 128 | approx = cv2.approxPolyDP(c, 0.1 * peri, True) 129 | (x, y, w, h) = cv2.boundingRect(approx) 130 | ar = h / float(w) 131 | 132 | # 筛选轮廓为四边形的目前轮廓 133 | # if y >= 60 and w >= 20 and w <= 60 and ar >= 1 and ar <= 2 and area > 700: 134 | if y >= 60 and ar > 0.5 and ar < 2: 135 | if is_choice_option(x, w, all_options_center_x) and area > 400: 136 | choice_option_cnts.append(c) 137 | elif not is_choice_option(x, w, all_options_center_x) and area > 100: 138 | question_number_cnts.append(c) 139 | return choice_option_cnts, question_number_cnts 140 | 141 | 142 | def is_choice_option(x, w, all_option_center_x): 143 | for center_x in all_option_center_x: 144 | if center_x > x and center_x < x + w: 145 | return True 146 | return False 147 | 148 | 149 | def get_cnt_center_x(cnts): 150 | """返回轮廓中心的x轴坐标 151 | 152 | Args: 153 | cnts (list): 轮廓列表 154 | 155 | Returns: 156 | [list]: 中心x轴坐标 157 | """ 158 | center_x = [] 159 | for cnt in cnts: 160 | (x, y, w, h) = cv2.boundingRect(cnt) 161 | center_x.append((2 * x + w) / 2) 162 | return center_x 163 | 164 | 165 | def get_choice_option_center_x(img): 166 | """ 识别所有未被填充的选择项的中心的x坐标 167 | 168 | Args: 169 | img ([type]): [description] 170 | 171 | Returns: 172 | [type]: [description] 173 | """ 174 | img = cv2.imread(img) 175 | ocr_reslut = pytesseract.image_to_data(img, output_type=Output.DICT, lang='chi_sim') 176 | 177 | choice_option_center_x = [] 178 | for i in range(len(ocr_reslut['text'])): 179 | text_i = ocr_reslut['text'][i] 180 | (x, y, w, _) = (ocr_reslut['left'][i], ocr_reslut['top'][i], ocr_reslut['width'][i], ocr_reslut['height'][i]) 181 | if y > 60 and ('A' in text_i or 'B' in text_i or 'C' in text_i or 'D' in text_i): 182 | choice_option_center_x.append((2 * x + w) / 2) 183 | return choice_option_center_x 184 | 185 | 186 | def get_answer_card_cnts(img): 187 | """ 获得答题卡的左右答题区域 188 | # findContours 函数详解：https://blog.csdn.net/laobai1015/article/details/76400725 189 | # approxPolyDP 多边形近似 https://blog.csdn.net/kakiebu/article/details/79824856 190 | 191 | Args: 192 | img ([type]): 图片 193 | Returns: 194 | [type]: 答题卡的左右答题区域轮廓 195 | """ 196 | 197 | # 检测图片中的最外围轮廓 198 | cnts, _ = cv2.findContours(img.copy(), cv2.RETR_EXTERNAL, 199 | cv2.CHAIN_APPROX_SIMPLE) 200 | # print("原始图片检测的轮廓总数：", len(cnts)) 201 | if len(cnts) == 0: 202 | return None 203 | 204 | # 提取的轮廓总数 205 | contour_size = 0 206 | # 检测到的左右答题区域轮廓 207 | answer_cnts = [] 208 | 209 | # 将轮廓按大小, 降序排序 210 | cnts = sorted(cnts, key=cv2.contourArea, reverse=True) 211 | for c in cnts: 212 | # arcLength 计算周长 213 | peri = cv2.arcLength(c, True) 214 | # print("轮廓周长：", peri) 215 | 216 | # 之前寻找到的轮廓可能是多边形，现在通过寻找近似轮廓，得到期望的四边形 217 | approx = cv2.approxPolyDP(c, 0.02 * peri, True) 218 | # print('原始轮廓的边数:', len(c), ', 近似轮廓的边数:', len(approx)) 219 | 220 | # 当近似轮廓为4时，代表是需要提取的矩形区域 221 | if len(approx) == 4: 222 | contour_size = contour_size + 1 223 | answer_cnts.append(approx) 224 | 225 | # 只提取答题卡中的最大两个轮廓 226 | if contour_size == ANSWER_CARD_SIZE: 227 | break 228 | 229 | answer_cnts = sort_contours(answer_cnts, method="left-to-right")[0] 230 | return answer_cnts 231 | 232 | 233 | def get_sub_answer_card_cnts(img_path): 234 | """ 获得答题卡的子区域 235 | # findContours 函数详解：https://blog.csdn.net/laobai1015/article/details/76400725 236 | # approxPolyDP 多边形近似 https://blog.csdn.net/kakiebu/article/details/79824856 237 | 238 | Args: 239 | img ([type]): 图片 240 | Returns: 241 | [type]: 答题卡的左右答题区域轮廓 242 | """ 243 | image = cv2.imread(img_path) 244 | gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) 245 | # warped_answer_image_1 = four_point_transform(gray, answer_contour_1.reshape(4, 2)) 246 | 247 | # 二值化 248 | thresh = cv2.threshold(gray, 0, 255, 249 | cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)[1] 250 | # 在二值图像中查找轮廓，包括内围、外围轮廓，但是检测到的轮廓不建立等级关系，彼此之间独立 251 | thresh_cnts, _ = cv2.findContours(thresh.copy(), cv2.RETR_LIST, 252 | cv2.CHAIN_APPROX_SIMPLE) 253 | 254 | cnt_size = 0 255 | sub_answer_cnts = [] 256 | if len(thresh_cnts) > 0: 257 | # 将轮廓按大小, 降序排序 258 | thresh_cnts = sorted(thresh_cnts, key=cv2.contourArea, reverse=True) 259 | for c in thresh_cnts: 260 | cnt_size = cnt_size + 1 261 | 262 | # arcLength 计算周长 263 | peri = cv2.arcLength(c, True) 264 | 265 | # 计算轮廓的边界框 266 | (x, y, w, h) = cv2.boundingRect(c) 267 | 268 | # 之前寻找到的轮廓可能是多边形，现在通过寻找近似轮廓，得到期望的四边形 269 | approx = cv2.approxPolyDP(c, 0.02 * peri, True) 270 | 271 | # 只提取近似轮廓为四边形的区域, 且轮廓长度大于指定长度 272 | # if len(approx) == 4 and w > ANSWER_CARD_MIN_WIDTH: 273 | 274 | # print("轮廓周长：", peri, '宽:', w) 275 | # print('原始轮廓的边数:', len(c), ', 近似轮廓的边数:', len(approx)) 276 | if w > ANSWER_CARD_MIN_WIDTH: 277 | sub_answer_cnts.append(approx) 278 | 279 | # 只处理前20个最大轮廓 280 | if cnt_size >= 20: 281 | break 282 | 283 | # 从上到下，将轮廓排序 284 | sub_answer_cnts = sort_contours(sub_answer_cnts, method="top-to-bottom")[0] 285 | return sub_answer_cnts 286 | 287 | 288 | def get_question_num_dict(image, question_number_cnts, ocr): 289 | """获取图片中所有的选择题的题序 290 | 291 | Args: 292 | image ([type]): 图片 293 | question_number_cnts ([type]): 图片中的所有的选择题的题序轮廓 294 | ocr ([type]): ocr识别工具 295 | 296 | Returns: 297 | [dict]: key: 题序, value: 题序轮廓的坐标 298 | """ 299 | question_num_dict = {} 300 | for question_number_cnt in question_number_cnts: 301 | peri = cv2.arcLength(question_number_cnt, True) 302 | approx = cv2.approxPolyDP(question_number_cnt, 0.1 * peri, True) 303 | (x, y, w, h) = cv2.boundingRect(approx) 304 | 305 | # ocr识别题型轮廓区域的文本 306 | text = ocr.ocr_for_single_line(image[y:y + h, x:x + w]) 307 | question_num = ''.join(text) 308 | question_num = question_num.replace('.', '') 309 | 310 | # 文本是否为数字 311 | if question_num.isdigit(): 312 | (x, y, w, h) = cv2.boundingRect(question_number_cnt) 313 | question_num_dict[int(question_num)] = (x, y, w, h) 314 | 315 | # 按照题序从小到大排序 316 | question_num_list = sorted(question_num_dict.items(), key=lambda item: item[0]) 317 | return dict(question_num_list) 318 | 319 | 320 | def get_choice_question_answer_index(image, choice_option_cnts, answer_option_cnts, question_number_cnts, ocr): 321 | """自动批改, 返回每道试题对应的答案索引. \ 322 | 注意：(1)用户可能没有填充答案 (2)选择题的答案数量可能大于1 323 | 324 | Args: 325 | choice_option_cnts (list): 试题的选择项轮廓 326 | answer_option_cnts (list): 用户填充的答案轮廓 327 | question_number_cnts (list): 试题的题序轮廓 328 | Returns: 329 | [dict]: key 题序, value 答案索引列表 330 | """ 331 | 332 | # 获取所有选择题的题序 333 | question_num_dict = get_question_num_dict(image, question_number_cnts, ocr) 334 | 335 | question_answer_dict = {} 336 | for key in question_num_dict.keys(): 337 | (num_x, num_y, num_w, num_h) = question_num_dict[key] 338 | num_center_x = (2 * num_x + num_w) / 2 339 | num_center_y = (2 * num_y + num_h) / 2 340 | 341 | # 获取同一行中，本题序右侧第一个题序的中心x坐标 342 | min_num_center_x = float("inf") # 无穷大 343 | for question_number_cnt in question_number_cnts: 344 | (x, y, w, h) = cv2.boundingRect(question_number_cnt) 345 | right_num_center_x = (2 * x + w) / 2 346 | if num_center_y > y and num_center_y < y + h and right_num_center_x > num_center_x and right_num_center_x < min_num_center_x: 347 | min_num_center_x = right_num_center_x 348 | # print(min_num_center_x) 349 | 350 | # 获取本题的全部答案轮廓的中心x坐标列表 351 | # 一道选择题题可能有多个答案， 所以answers_center_x为列表 352 | answers_center_x = [] 353 | for answer_option_cnt in answer_option_cnts: 354 | (x, y, w, h) = cv2.boundingRect(answer_option_cnt) 355 | answer_cnt_center_x = (2 * x + w) / 2 356 | if num_center_y > y and num_center_y < y + h and answer_cnt_center_x > num_center_x and answer_cnt_center_x < min_num_center_x: 357 | answers_center_x.append(answer_cnt_center_x) 358 | # print('answers_center_x', answers_center_x) 359 | 360 | # 获取本题的全部选择项轮廓 361 | question_choice_option_cnts = [] 362 | for choice_option_cnt in choice_option_cnts: 363 | # print(len(question_choice_option_cnts)) 364 | (x, y, w, h) = cv2.boundingRect(choice_option_cnt) 365 | choice_option_center_x = (2 * x + w) / 2 366 | if num_center_y > y and num_center_y < y + h and choice_option_center_x > num_center_x and choice_option_center_x < min_num_center_x: 367 | question_choice_option_cnts.append(choice_option_cnt) 368 | 369 | question_choice_option_cnts, _ = sort_contours(question_choice_option_cnts, 'left-to-right') 370 | # print('question_choice_option_cnts', len(question_choice_option_cnts)) 371 | 372 | # 答案列表 373 | answer_indexes = [] 374 | # 答案索引 375 | answer_index = 0 376 | for choice_option_cnt in question_choice_option_cnts: 377 | answer_index = answer_index + 1 378 | (x, y, w, h) = cv2.boundingRect(choice_option_cnt) 379 | # print((x, y, w, h), answers_center_x) 380 | for answer_center_x in answers_center_x: 381 | if answer_center_x > x and answer_center_x < x + w: 382 | answer_indexes.append(answer_index) 383 | break 384 | question_answer_dict[key] = answer_indexes 385 | 386 | # 返回每道试题对应的答案索引 387 | question_answer_dict = sorted(question_answer_dict.items(), key=lambda item: item[0]) 388 | return dict(question_answer_dict) 389 | 390 | 391 | def is_choice_question(img): 392 | """判断当前图片是否属于选择题 393 | 394 | Args: 395 | image_path ([type]): 图片 396 | 397 | Returns: 398 | [boolean]: false 不是 true 是 399 | """ 400 | ocr_result = pytesseract.image_to_data(img, output_type=Output.DICT, lang='chi_sim') 401 | ocr_text = ocr_result['text'] 402 | return '[A]' in ocr_text or '[B]' in ocr_text or '[C]' in ocr_text or '[D]' in ocr_text 403 | -------------------------------------------------------------------------------- /detection_exam_num.py: -------------------------------------------------------------------------------- 1 | ''' 2 | Author: wangxin 3 | Date: 2021-05-25 10:31:01 4 | LastEditTime: 2021-07-01 13:55:35 5 | LastEditors: Please set LastEditors 6 | Description: 检测考号 7 | ''' 8 | 9 | # coding=utf-8 10 | import cv2 11 | import numpy as np 12 | from PIL import Image 13 | from utils import sort_contours 14 | 15 | 16 | def get_exam_num_area(image_path): 17 | """ 获取图片中待检测的考号填充区域 18 | 19 | Args: 20 | image_path (String): 图片地址 21 | 22 | Returns: 23 | [type]: [description] 24 | """ 25 | image = Image.open(image_path) 26 | image_width = image.width 27 | 28 | img = cv2.imread(image_path) 29 | gray_img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) 30 | threshold_img = cv2.threshold(gray_img, 0, 255, cv2.THRESH_BINARY_INV | cv2.THRESH_OTSU)[1] 31 | 32 | # 先腐蚀与膨胀, 高亮化学生填充的考号 33 | kernel = np.ones((9, 9), np.uint8) 34 | erode_img = cv2.erode(threshold_img, kernel, iterations=1) 35 | kernel = np.ones((9, 9), np.uint8) 36 | dilate_img = cv2.dilate(erode_img, kernel, iterations=1) 37 | 38 | # 学生填充的考号，最左边边缘的x轴坐标 39 | exam_number_left_x = float("inf") 40 | # 学生填充的考号，最右边边缘的x轴坐标 41 | exam_number_right_x = 0 42 | cnts, _ = cv2.findContours(dilate_img.copy(), cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE) 43 | for c in cnts: 44 | (x, y, w, h) = cv2.boundingRect(c) 45 | if x > image_width / 2: 46 | if x < exam_number_left_x: 47 | exam_number_left_x = x 48 | if x + w > exam_number_right_x: 49 | exam_number_right_x = x + w 50 | 51 | # 通过x轴坐标，缩小待检测区域的范围 52 | threshold_img = threshold_img[:, exam_number_left_x - 15:exam_number_right_x + 15] 53 | 54 | # 再通过检测图片中面积最大的轮廓（考号手写区域, 而不是填充区域）, 进一步缩小范围 55 | cnts, _ = cv2.findContours(threshold_img, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE) 56 | cnts = sorted(cnts, key=cv2.contourArea, reverse=True) 57 | mix_y = None 58 | num_card_cnt = None 59 | for c in cnts: 60 | peri = cv2.arcLength(c, True) 61 | approx = cv2.approxPolyDP(c, 0.06 * peri, True) 62 | (x, y, w, h) = cv2.boundingRect(approx) 63 | if len(approx) == 4: 64 | cv2.imwrite('out/num_card.jpg', threshold_img[y:y + h, x:x + w]) 65 | num_card_cnt = c 66 | mix_y = y + h 67 | break 68 | 69 | threshold_img = threshold_img[mix_y:, :] 70 | return threshold_img, num_card_cnt 71 | 72 | 73 | def get_exam_num_height(img): 74 | """ 获取考号填充区域, 行中心与行中心的y轴坐标间隔 75 | 76 | Args: 77 | img ([type]): 图片 78 | 79 | Returns: 80 | [float]: 行中心与行中心的y轴坐标间隔 81 | """ 82 | 83 | # 膨胀 84 | kernel = np.ones((5, 5), np.uint8) 85 | dilate_img = cv2.dilate(img, kernel, iterations=1) 86 | 87 | # 第一行待填充考号的中心的x轴坐标 88 | first_line_center_y = None 89 | # 第二行待填充考号的中心的x轴坐标 90 | second_line_center_y = None 91 | # 第一行待填充考号的底部边缘的y坐标 92 | first_line_bottom_y = None 93 | cnts, _ = cv2.findContours(dilate_img, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE) 94 | cnts, _ = sort_contours(cnts, 'top-to-bottom') 95 | for c in cnts: 96 | peri = cv2.arcLength(c, True) 97 | approx = cv2.approxPolyDP(c, 0.06 * peri, True) 98 | (x, y, w, h) = cv2.boundingRect(approx) 99 | center_y = (2 * y + h) / 2 100 | 101 | if h > 10: 102 | if first_line_center_y is None: 103 | first_line_center_y = center_y 104 | first_line_bottom_y = y + h 105 | 106 | if center_y > first_line_bottom_y and second_line_center_y is None: 107 | second_line_center_y = center_y 108 | break 109 | 110 | print(type(second_line_center_y)) 111 | return second_line_center_y - first_line_center_y 112 | 113 | 114 | def detection_exam_num(image_path): 115 | """ 识别图片中学生填充的考号 116 | 117 | Args: 118 | image_path (String): 图片地址 119 | 120 | Returns: 121 | [list]: 识别的考号结果 122 | """ 123 | # 获取图片中考号填充区域范围 124 | thresh_img, _ = get_exam_num_area(image_path) 125 | 126 | # 获取考号填充区域, 每2行的中心y轴坐标间隔 127 | line_y_height = get_exam_num_height(thresh_img) 128 | 129 | # 腐蚀与膨胀 130 | kernel = np.ones((9, 9), np.uint8) 131 | erode_img = cv2.erode(thresh_img, kernel, iterations=1) 132 | kernel = np.ones((9, 9), np.uint8) 133 | dilate_img = cv2.dilate(erode_img, kernel, iterations=1) 134 | 135 | # 学生填充考号的识别结果 136 | num_card = [] 137 | cnts, _ = cv2.findContours(dilate_img, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE) 138 | cnts, boundingBoxes = sort_contours(cnts, 'left-to-right') 139 | for c in cnts: 140 | peri = cv2.arcLength(c, True) 141 | approx = cv2.approxPolyDP(c, 0.06 * peri, True) 142 | (x, y, w, h) = cv2.boundingRect(approx) 143 | num_card.append(int(y / line_y_height)) 144 | return num_card 145 | 146 | 147 | if __name__ == '__main__': 148 | num_card = detection_exam_num('out/sub_answer_card_0.jpg') 149 | print('num_card: ', num_card) 150 | -------------------------------------------------------------------------------- /main.py: -------------------------------------------------------------------------------- 1 | ''' 2 | Author: wangxin 3 | Date: 2021-05-25 10:29:58 4 | LastEditTime: 2021-07-01 14:12:15 5 | LastEditors: Please set LastEditors 6 | Description: In User Settings Edit 7 | ''' 8 | 9 | from detection_choice_question import get_answer_card_cnts, get_sub_answer_card_cnts, detection_choice_question 10 | from detection_exam_num import detection_exam_num 11 | from settings import TITLE_NUM 12 | from utils import get_init_process_img, capture_img, ocr_single_line_img 13 | from cnocr import CnOcr 14 | 15 | 16 | def demo(origin_image_path): 17 | # 获取答题卡左右区域 18 | image = get_init_process_img(origin_image_path) 19 | answer_cnts = get_answer_card_cnts(image) 20 | answer_card_images_path = [] 21 | if len(answer_cnts) > 0: 22 | len_answer_cnts = 0 23 | for c in answer_cnts: 24 | len_answer_cnts = len_answer_cnts + 1 25 | answer_card_image_path = 'out/answer_card_' + str(len_answer_cnts) + '.jpg' 26 | answer_card_images_path.append(answer_card_image_path) 27 | capture_img(origin_image_path, answer_card_image_path, c) 28 | print('答题卡左右区域切分结果：', answer_card_images_path) 29 | 30 | # 将答题卡切分为一道道试题 31 | sub_answer_card_images_path = [] 32 | sub_answer_cnt_szie = 0 33 | for answer_card_image in answer_card_images_path: 34 | sub_answer_cnts = get_sub_answer_card_cnts(answer_card_image) 35 | if len(sub_answer_cnts) > 1: 36 | sub_answer_cnts = sub_answer_cnts[1:len(sub_answer_cnts)] 37 | 38 | if len(sub_answer_cnts) > 0: 39 | for c in sub_answer_cnts: 40 | sub_answer_card_image_path = 'out/sub_answer_card_' + str(sub_answer_cnt_szie) + '.jpg' 41 | sub_answer_card_images_path.append(sub_answer_card_image_path) 42 | capture_img(answer_card_image, sub_answer_card_image_path, c) 43 | sub_answer_cnt_szie = sub_answer_cnt_szie + 1 44 | print('试题切分结果：', sub_answer_card_images_path) 45 | 46 | # 获取每个大标题的索引 47 | ocr = CnOcr() 48 | title_index = [] 49 | for img in sub_answer_card_images_path: 50 | res = ocr_single_line_img(img, ocr) 51 | if len(res) > 0 and res[0] in TITLE_NUM: 52 | title_index.append(sub_answer_card_images_path.index(img)) 53 | print('每道大题的起始图片索引: ', title_index) 54 | 55 | # 学生考号自动识别 56 | num_card = detection_exam_num(sub_answer_card_images_path[0]) 57 | print('学生考号: ', num_card) 58 | 59 | # 选择题自动识别与批改 60 | question_answer_dict = detection_choice_question(sub_answer_card_images_path, ocr) 61 | print('每道选择题答案（key 题序, value: 对应题序的答案列表）：', question_answer_dict) 62 | 63 | 64 | if __name__ == '__main__': 65 | demo('pic/answer_card1.jpg') 66 | -------------------------------------------------------------------------------- /pic/answer_card.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Wasim37/answer_card_detection/b45b2c4fc06f757a78eb0b16188b939bdb17267f/pic/answer_card.jpg -------------------------------------------------------------------------------- /pic/choice_question_detection_1.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Wasim37/answer_card_detection/b45b2c4fc06f757a78eb0b16188b939bdb17267f/pic/choice_question_detection_1.jpg -------------------------------------------------------------------------------- /pic/choice_question_detection_2.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Wasim37/answer_card_detection/b45b2c4fc06f757a78eb0b16188b939bdb17267f/pic/choice_question_detection_2.png -------------------------------------------------------------------------------- /pic/choice_question_detection_3.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Wasim37/answer_card_detection/b45b2c4fc06f757a78eb0b16188b939bdb17267f/pic/choice_question_detection_3.png -------------------------------------------------------------------------------- /pic/choice_question_detection_4.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Wasim37/answer_card_detection/b45b2c4fc06f757a78eb0b16188b939bdb17267f/pic/choice_question_detection_4.png -------------------------------------------------------------------------------- /pic/choice_question_detection_5.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Wasim37/answer_card_detection/b45b2c4fc06f757a78eb0b16188b939bdb17267f/pic/choice_question_detection_5.png -------------------------------------------------------------------------------- /pic/num_card_detection_1.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Wasim37/answer_card_detection/b45b2c4fc06f757a78eb0b16188b939bdb17267f/pic/num_card_detection_1.png -------------------------------------------------------------------------------- /pic/num_card_detection_2.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Wasim37/answer_card_detection/b45b2c4fc06f757a78eb0b16188b939bdb17267f/pic/num_card_detection_2.png -------------------------------------------------------------------------------- /pic/num_card_detection_3.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Wasim37/answer_card_detection/b45b2c4fc06f757a78eb0b16188b939bdb17267f/pic/num_card_detection_3.png -------------------------------------------------------------------------------- /pic/special_choice_question.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Wasim37/answer_card_detection/b45b2c4fc06f757a78eb0b16188b939bdb17267f/pic/special_choice_question.png -------------------------------------------------------------------------------- /pic/special_num_card_1.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Wasim37/answer_card_detection/b45b2c4fc06f757a78eb0b16188b939bdb17267f/pic/special_num_card_1.png -------------------------------------------------------------------------------- /pic/special_num_card_2.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Wasim37/answer_card_detection/b45b2c4fc06f757a78eb0b16188b939bdb17267f/pic/special_num_card_2.png -------------------------------------------------------------------------------- /pic/sub_answer_card_1.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Wasim37/answer_card_detection/b45b2c4fc06f757a78eb0b16188b939bdb17267f/pic/sub_answer_card_1.jpg -------------------------------------------------------------------------------- /pic/sub_answer_card_2.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Wasim37/answer_card_detection/b45b2c4fc06f757a78eb0b16188b939bdb17267f/pic/sub_answer_card_2.jpg -------------------------------------------------------------------------------- /pic/sub_answer_card_3.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Wasim37/answer_card_detection/b45b2c4fc06f757a78eb0b16188b939bdb17267f/pic/sub_answer_card_3.jpg -------------------------------------------------------------------------------- /pic/sub_answer_card_4.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Wasim37/answer_card_detection/b45b2c4fc06f757a78eb0b16188b939bdb17267f/pic/sub_answer_card_4.jpg -------------------------------------------------------------------------------- /pic/sub_answer_card_5.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Wasim37/answer_card_detection/b45b2c4fc06f757a78eb0b16188b939bdb17267f/pic/sub_answer_card_5.jpg -------------------------------------------------------------------------------- /pic/sub_answer_card_6.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Wasim37/answer_card_detection/b45b2c4fc06f757a78eb0b16188b939bdb17267f/pic/sub_answer_card_6.jpg -------------------------------------------------------------------------------- /pic/sub_answer_card_7.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Wasim37/answer_card_detection/b45b2c4fc06f757a78eb0b16188b939bdb17267f/pic/sub_answer_card_7.jpg -------------------------------------------------------------------------------- /pic/sub_answer_card_8.jpg: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Wasim37/answer_card_detection/b45b2c4fc06f757a78eb0b16188b939bdb17267f/pic/sub_answer_card_8.jpg -------------------------------------------------------------------------------- /requirements.txt: -------------------------------------------------------------------------------- 1 | opencv-python==4.5.2.52 2 | imutils==0.5.4 3 | pytesseract==0.3.7 4 | -------------------------------------------------------------------------------- /settings.py: -------------------------------------------------------------------------------- 1 | ''' 2 | Author: wangxin 3 | Date: 2021-05-25 10:31:52 4 | LastEditTime: 2021-07-01 17:04:28 5 | LastEditors: Please set LastEditors 6 | Description: In User Settings Edit 7 | ''' 8 | 9 | # coding=utf-8 10 | 11 | # 图片中的答题卡框数量. 比如一张图片可以划分为左右2个答题卡框 12 | ANSWER_CARD_SIZE = 2 13 | 14 | # 答题卡框的最小宽度 15 | ANSWER_CARD_MIN_WIDTH = 1200 16 | 17 | # 大标题序号 18 | TITLE_NUM = ['一', '二', '三', '四', '五', '六', '七', '八', '九', '十', '十一', '十二', '十三', '十四', '十五', '十六', '十七'] 19 | 20 | # 识别左上角大标题序号, 识别范围的宽 21 | TITLE_TOP_LEFT_CORNER_WIDTH = 50 22 | 23 | # 识别左上角大标题序号, 识别范围的高 24 | TITLE_TOP_LEFT_CORNER_HEIGTH = 65 25 | -------------------------------------------------------------------------------- /test/test_cnocr.py: -------------------------------------------------------------------------------- 1 | ''' 2 | Author: your name 3 | Date: 2021-06-02 17:55:10 4 | LastEditTime: 2021-06-03 15:40:43 5 | LastEditors: Please set LastEditors 6 | Description: In User Settings Edit 7 | FilePath: \teaching-answer-card-tool\test.py 8 | ''' 9 | import cv2 10 | from cnocr import CnOcr 11 | 12 | 13 | if __name__ == '__main__': 14 | img = cv2.imread('out/sub_answer_card_1.jpg') 15 | ocr = CnOcr() 16 | res = ocr.ocr_for_single_line(img) 17 | print(res) 18 | -------------------------------------------------------------------------------- /test/test_pytesseract.py: -------------------------------------------------------------------------------- 1 | ''' 2 | Author: your name 3 | Date: 2021-06-02 17:55:10 4 | LastEditTime: 2021-06-03 15:40:43 5 | LastEditors: Please set LastEditors 6 | Description: In User Settings Edit 7 | FilePath: \teaching-answer-card-tool\test.py 8 | ''' 9 | import cv2 10 | import pytesseract 11 | from utils import get_init_process_img 12 | from pytesseract import Output 13 | from PIL import Image 14 | from PIL import ImageDraw 15 | from PIL import ImageFont 16 | import numpy as np 17 | 18 | # https://livezingy.com/pytesseract-image_to_data_locate_text/ 19 | # pytesseract.pytesseract.tesseract_cmd = r'D:\ProgramData\Tesseract-OCR\tesseract.exe' 20 | 21 | 22 | def recoText(im): 23 | """ 24 | 识别字符并返回所识别的字符及它们的坐标 25 | :param im: 需要识别的图片 26 | :return data: 字符及它们在图片的位置 27 | """ 28 | data = {} 29 | # im = get_init_process_img(im) 30 | # d = pytesseract.image_to_string(im, output_type=Output.DICT, lang='chi_sim') 31 | # print(d['text']) 32 | 33 | dd = pytesseract.image_to_string(im, output_type=Output.DICT, lang="eng") 34 | print(dd['text']) 35 | # print(d['text']) 36 | # for i in range(len(d['text'])): 37 | # print(d['text'][i]) 38 | # if 0 < len(d['text'][i]): 39 | # (x, y, w, h) = (d['left'][i], d['top'][i], d['width'][i], d['height'][i]) 40 | # data[d['text'][i]] = ([d['left'][i], d['top'][i], d['width'][i], d['height'][i]]) 41 | 42 | # cv2.rectangle(im, (x, y), (x + w, y + h), (255, 0, 0), 1) 43 | # # 使用cv2.putText不能显示中文，需要使用下面的代码代替 44 | # # cv2.putText(im, d['text'][i], (x, y-8), cv2.FONT_HERSHEY_SIMPLEX, 0.3, (255, 0, 0), 1) 45 | 46 | # pilimg = Image.fromarray(im) 47 | # draw = ImageDraw.Draw(pilimg) 48 | # # 参数1：字体文件路径，参数2：字体大小 49 | # # Hiragino Sans GB.ttc 为mac下的简体中文 50 | # font = ImageFont.truetype("Hiragino Sans GB.ttc", 15, encoding="utf-8") 51 | # # 参数1：打印坐标，参数2：文本，参数3：字体颜色，参数4：字体 52 | # draw.text((x, y - 10), d['text'][i], (255, 0, 0), font=font) 53 | # im = cv2.cvtColor(np.array(pilimg), cv2.COLOR_RGB2BGR) 54 | 55 | # cv2.imshow("recoText", im) 56 | return data 57 | 58 | 59 | def recognize_text(image): 60 | # 边缘保留滤波 去噪 61 | blur = cv2.pyrMeanShiftFiltering(image, sp=8, sr=60) 62 | cv2.imshow('dst', blur) 63 | # 灰度图像 64 | gray = cv2.cvtColor(blur, cv2.COLOR_BGR2GRAY) 65 | # 二值化 设置阈值 自适应阈值的话 黄色的4会提取不出来 66 | ret, binary = cv2.threshold(gray, 185, 255, cv2.THRESH_BINARY_INV) 67 | print(f'二值化设置的阈值：{ret}') 68 | cv2.imshow('binary', binary) 69 | # 逻辑运算 让背景为白色 字体为黑 便于识别 70 | cv2.bitwise_not(binary, binary) 71 | cv2.imshow('bg_image', binary) 72 | # 识别 73 | test_message = Image.fromarray(binary) 74 | text = pytesseract.image_to_string(test_message) 75 | print(f'识别结果：{text}') 76 | 77 | 78 | if __name__ == '__main__': 79 | # img = cv2.imread('out/sub_answer_card_3.jpg') 80 | 81 | img = cv2.imread('20210623184007.jpg') 82 | 83 | # 转灰度 84 | # img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) 85 | # ret, img = cv2.threshold(img, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU) # 二值化 86 | 87 | # height, width, deep = img.shape # cropImg是从图片里截取的,只包含一行数字 88 | # gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY) # 转灰度图 89 | # dst = np.zeros((height, width, 1), np.uint8) 90 | # for i in range(0, height): # 反相 转白底黑字 91 | # for j in range(0, width): 92 | # grayPixel = gray[i, j] 93 | # dst[i, j] = 255 - grayPixel 94 | # ret, canny = cv2.threshold(dst, 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU) # 二值化 95 | 96 | 97 | # img = cv2.imread('out/test.png') 98 | # cv2.imshow("src", img) 99 | # data = recoText(img) 100 | recognize_text(img) 101 | 102 | cv2.waitKey(0) 103 | cv2.destroyAllWindows() 104 | -------------------------------------------------------------------------------- /utils.py: -------------------------------------------------------------------------------- 1 | import numpy as np 2 | import cv2 3 | import matplotlib.pyplot as plt 4 | from settings import TITLE_TOP_LEFT_CORNER_WIDTH, TITLE_TOP_LEFT_CORNER_HEIGTH 5 | from imutils import auto_canny 6 | 7 | 8 | def order_points(pts): 9 | """4边形4点排序函数 10 | 11 | Args: 12 | pts ([type]): 4边形任意顺序的4个顶点 13 | 14 | Returns: 15 | [type]: 按照一定顺序的4个顶点 16 | """ 17 | 18 | rect = np.zeros((4, 2), dtype="float32") # 按照左上、右上、右下、左下顺序初始化坐标 19 | 20 | s = pts.sum(axis=1) # 计算点xy的和 21 | rect[0] = pts[np.argmin(s)] # 左上角的点的和最小 22 | rect[2] = pts[np.argmax(s)] # 右下角的点的和最大 23 | 24 | diff = np.diff(pts, axis=1) # 计算点xy之间的差 25 | rect[1] = pts[np.argmin(diff)] # 右上角的差最小 26 | rect[3] = pts[np.argmax(diff)] # 左下角的差最小 27 | return rect # 返回4个顶点的顺序 28 | 29 | 30 | def four_point_transform(image, pts): 31 | """4点变换 32 | 33 | Args: 34 | image ([type]): 原始图像 35 | pts ([type]): 4个顶点 36 | 37 | Returns: 38 | [type]: 变换后的图像 39 | """ 40 | 41 | rect = order_points(pts) # 获得一致的顺序的点并分别解包他们 42 | (tl, tr, br, bl) = rect 43 | 44 | # 计算新图像的宽度(x) 45 | widthA = np.sqrt(((br[0] - bl[0])**2) + ((br[1] - bl[1])**2)) # 右下和左下之间距离 46 | widthB = np.sqrt(((tr[0] - tl[0])**2) + ((tr[1] - tl[1])**2)) # 右上和左上之间距离 47 | maxWidth = max(int(widthA), int(widthB)) # 取大者 48 | 49 | # 计算新图像的高度(y) 50 | heightA = np.sqrt(((tr[0] - br[0])**2) + ((tr[1] - br[1])**2)) # 右上和右下之间距离 51 | heightB = np.sqrt(((tl[0] - bl[0])**2) + ((tl[1] - bl[1])**2)) # 左上和左下之间距离 52 | maxHeight = max(int(heightA), int(heightB)) 53 | 54 | # 有了新图像的尺寸, 构造透视变换后的顶点集合 55 | dst = np.array( 56 | [ 57 | [0, 0], # -------------------------左上 58 | [maxWidth - 1, 0], # --------------右上 59 | [maxWidth - 1, maxHeight - 1], # --右下 60 | [0, maxHeight - 1] 61 | ], # ------------左下 62 | dtype="float32") 63 | 64 | M = cv2.getPerspectiveTransform(rect, dst) # 计算透视变换矩阵 65 | warped = cv2.warpPerspective(image, M, (maxWidth, maxHeight)) # 执行透视变换 66 | 67 | return warped # 返回透视变换后的图像 68 | 69 | 70 | def sort_contours(cnts, method="left-to-right"): 71 | """轮廓排序 72 | 73 | Args: 74 | cnts ([type]): 轮廓 75 | method (str, optional): 排序方式. Defaults to "left-to-right". 76 | 77 | Returns: 78 | [type]: 排序好的轮廓 79 | """ 80 | 81 | if cnts is None or len(cnts) == 0: 82 | return [], [] 83 | 84 | # 初始化逆序标志和排序索引 85 | reverse = False 86 | i = 0 87 | 88 | # 是否需逆序处理 89 | if method == "right-to-left" or method == "bottom-to-top": 90 | reverse = True 91 | 92 | # 是否需要按照y坐标函数 93 | if method == "top-to-bottom" or method == "bottom-to-top": 94 | i = 1 95 | 96 | # 构造包围框列表，并从上到下对它们进行排序 97 | boundingBoxes = [cv2.boundingRect(c) for c in cnts] 98 | (cnts, boundingBoxes) = zip(*sorted( 99 | zip(cnts, boundingBoxes), key=lambda b: b[1][i], reverse=reverse)) 100 | 101 | # 返回已排序的轮廓线和边框列表 102 | return cnts, boundingBoxes 103 | 104 | 105 | def get_init_process_img(img_path): 106 | """ 107 | 对图片进行初始化处理，包括灰度，高斯模糊，腐蚀，膨胀和边缘检测等 108 | :param roi_img: ndarray 109 | :return: ndarray 110 | """ 111 | image = cv2.imread(img_path) 112 | # 转灰度 113 | gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY) 114 | # 高斯模糊 115 | blurred = cv2.GaussianBlur(gray, (5, 5), 0) 116 | 117 | # 腐蚀erode与膨胀dilate 118 | # kernel = np.ones((3, 3), np.uint8) 119 | # blurred = cv2.erode(blurred, kernel, iterations=1) # 腐蚀 120 | # blurred = cv2.dilate(blurred, kernel, iterations=2) # 膨胀 121 | # blurred = cv2.erode(blurred, kernel, iterations=1) # 腐蚀 122 | # blurred = cv2.dilate(blurred, kernel, iterations=2) # 膨胀 123 | 124 | # 边缘检测 125 | # edged = cv2.Canny(blurred, 75, 200) 126 | edged = auto_canny(blurred) 127 | return edged 128 | 129 | 130 | def capture_img(origin_image_path, target_image_path, contour): 131 | """根据轮廓截取图片 132 | 133 | Args: 134 | origin_image_path ([type]): 原始图片路径 135 | target_image_path ([type]): 目标图片路径 136 | contour ([type]): 截取轮廓 137 | 138 | Returns: 139 | [type]: [description] 140 | """ 141 | # 根据轮廓或者坐标 142 | x, y, w, h = cv2.boundingRect(contour) 143 | # 截图 144 | image = cv2.imread(origin_image_path) 145 | cv2.imwrite(target_image_path, image[y:y + h, x:x + w]) 146 | 147 | 148 | def save_img_by_cnts(save_image_path, image_size, cnts): 149 | """通过提取的轮廓绘制图片并保存 150 | 151 | Args: 152 | save_image_path ([type]): 图片存储路径 153 | image ([type]): 绘制的图片尺寸, 长与宽 154 | cnts ([type]): 轮廓列表 155 | """ 156 | black_background = np.ones(image_size, np.uint8) * 0 157 | cv2.drawContours(black_background, cnts, -1, (255, 255, 255), 2) 158 | plt.figure(figsize=(10, 5)) 159 | plt.imshow(black_background) 160 | plt.axis('off') 161 | plt.savefig(save_image_path) 162 | 163 | 164 | def ocr_single_line_img(image_path, ocr): 165 | """ocr识别图片 166 | 167 | Args: 168 | origin_image_path ([type]): 原始图片路径 169 | ocr ([type]): ocr 170 | 171 | Returns: 172 | [type]: [description] 173 | """ 174 | 175 | image = cv2.imread(image_path) 176 | res = ocr.ocr_for_single_line(image[0:TITLE_TOP_LEFT_CORNER_WIDTH, 0:TITLE_TOP_LEFT_CORNER_HEIGTH]) 177 | if len(res) > 0 and res[0] == '-': 178 | res[0] = '一' 179 | return res 180 | 181 | --------------------------------------------------------------------------------