zoukankan      html  css  js  c++  java
  • ocr 文字区域检测及识别

    ocr 文字区域检测及识别

    # coding=utf-8
    
    from PIL import Image, ImageFilter, ImageEnhance
    from skimage.filters import threshold_otsu
    import skimage.morphology as sm
    from skimage.measure import regionprops
    import matplotlib.pyplot as plt
    import numpy as np
    import pytesseract
    import re
    import os
    import time
    import logging
    logging.basicConfig(level=logging.INFO, format="%(message)s", filename='train_output.log')

    class ShopCert(object): def cut_region(self, img): """ 先按规则缩小搜索范围 """ w, h = img.size if h<1200: factor = max(1, 1600.0/h) newsize = int(w*factor), int(h*factor) img = img.resize(newsize, Image.ANTIALIAS) if w<h: box = (w*0.4, h*0.18, w*0.96, h*0.6) else: box = (w*0.1, h*0.18, w*0.96, h*0.9) return img.crop(box) def detect_text(self, img): """ 检测字符区域 """ imgM = np.array(img.convert('L')) imgM = 1 * (imgM < threshold_otsu(imgM)) imgM = sm.binary_closing(imgM, np.ones((5, 10))) imgM = sm.remove_small_objects(imgM, 600) label_img = sm.label(imgM) imgList = [] for region in regionprops(label_img): minr, minc, maxr, maxc = region.bbox w, h = (maxc-minc), (maxr-minr) if h > w * 0.2: continue box = minc-5, minr-3, maxc+5, maxr+3 imgList.append(img.crop(box)) return imgList def clear_noise(self, box): """ 降噪处理 """ box = box.convert('L') # box = box.point(lambda x: 0 if x<50 else x) box = box.point(lambda x: 200 if x>200 else x) box = ImageEnhance.Contrast(box).enhance(2.5) return box def predict(self, fname, lang='eng'): """ ocr 识别 """ img = Image.open(fname) # 先大致缩小范围 region = self.cut_region(img) # 候选字符区域 # region = self.clear_noise(region) boxList = self.detect_text(region) # 遍历识别 for box in boxList: box = self.clear_noise(box) w, h = box.size if float(w)/h > 12.5: res = pytesseract.image_to_string(box, lang='chi_sim', config='-psm 7') else: res = pytesseract.image_to_string(box, lang='eng', config='-psm 7') res = re.sub('s', '', res) # 去除中间空白 res = re.findall(r'[0-9][A-Z0-9]{13,20}', res) # 13-20位 for line in res: line = line.strip() if line.find(u'')>1: continue print 'line', line if len(line)> 14: box.save('img/clearNoise/%s_%s.jpg' % (fname.split('/')[-1].split('.')[0], line)) return line else: print 'error line', line return 'error' def show_pic(path='img/origin2/'): fnames = [os.path.join(path, fname) for fname in os.listdir(path)] for i, fname in enumerate(fnames, 0): print fname img = Image.open(fname) # img.save('./tesseract-train/cert.normal.exp%d.ttf' % i) img = ImageEnhance.Contrast(img).enhance(2.0) img = img.filter(ImageFilter.MedianFilter).convert('L') plt.figure(figsize=(10, 12), dpi=300) plt.imshow(img, plt.cm.gray) plt.title(fname.split('/')[-1]+'_%d' % i) plt.show()if __name__ == '__main__': test = ShopCert() path = 'img/origin2/' fnames = [os.path.join(path, fname) for fname in os.listdir(path) if fname.endswith('jpg')] fnames.sort() arguments = 'mode: L; enhance:2.0; h:0.5; dh:0.15' logging.info('%s' % arguments) logging.info("%s: %s" % ('imgname', 'result')) start_time = time.time() cnt = 0 for idx, fname in enumerate(fnames, 1): print idx, fname y_true = fname.split('/')[-1].split('_')[0] y_pred = test.predict(fname) if y_true == y_pred: cnt +=1 print fname else: print '***'*20 print 'error' logging.info("%s: %s" % (fname, y_pred)) print 'y_true', y_true print 'y_pred', y_pred acc = float(cnt)/idx print acc, cnt print '=='*20, idx logging.info('%.3f %d/%d' % (acc, cnt, idx)) print 'cost time: ', time.time()-start_time logging.info('accuracy: %.2f' % acc)
  • 相关阅读:
    线性dp打鼹鼠
    区间dp(能量项链)
    留言板
    友链
    CodeForces 1270I Xor on Figures
    agc020_d Min Max Repetition
    Codeforces 1270G Subset with Zero Sum
    CodeForecs 1270E Divide Points
    MybatisPlus学习笔记10:公共字段自动填充
    MybatisPlus学习笔记9:自定义全局操作
  • 原文地址:https://www.cnblogs.com/jkmiao/p/6814416.html
Copyright © 2011-2022 走看看