sub_ocr.py 6.2 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195
  1. # This Python file uses the following encoding: utf-8
  2. # @author runhey
  3. # github https://github.com/runhey
  4. import cv2
  5. import re
  6. from datetime import timedelta
  7. from module.ocr.ppocr import TextSystem
  8. from module.exception import ScriptError
  9. from module.base.utils import area_pad, crop, float2str
  10. from module.ocr.base_ocr import BaseCor, OcrMode, OcrMethod
  11. from module.ocr.utils import merge_area
  12. from module.logger import logger
  13. class Full(BaseCor):
  14. """
  15. 这个类适用于大ROI范围的文本识别。可以支持多条文本识别, 默认不支持竖方向的文本识别
  16. """
  17. def after_process(self, result):
  18. return result
  19. def ocr_full(self, image, keyword: str=None) -> tuple:
  20. """
  21. 检测整个图片的文本,并对结果进行过滤。返回的是匹配到的keyword的左边。如果没有匹配到返回(0, 0, 0, 0)
  22. :param image:
  23. :param keyword:
  24. :return:
  25. """
  26. if keyword is None:
  27. keyword = self.keyword
  28. boxed_results = self.detect_and_ocr(image)
  29. if not boxed_results:
  30. return 0, 0, 0, 0
  31. index_list = self.filter(boxed_results, keyword)
  32. # 如果一个都没有匹配到
  33. if not index_list:
  34. return 0, 0, 0, 0
  35. # 如果匹配到了多个,则合并所有的坐标,返回合并后的坐标
  36. if len(index_list) > 1:
  37. area_list = [(
  38. boxed_results[index].box[0, 0], # x
  39. boxed_results[index].box[0, 1], # y
  40. boxed_results[index].box[1, 0] - boxed_results[index].box[0, 0], # width
  41. boxed_results[index].box[2, 1] - boxed_results[index].box[0, 1], # height
  42. ) for index in index_list]
  43. box = merge_area(area_list)
  44. self.area = box[0]+self.roi[0], box[1]+self.roi[1], box[2], box[3]
  45. else:
  46. box = boxed_results[index_list[0]].box
  47. self.area = box[0, 0]+self.roi[0], box[0, 1]+self.roi[1], box[1, 0] - box[0, 0], box[2, 1] - box[0, 1]
  48. logger.info(f"OCR [{self.name}: {keyword}] detected in {self.area}")
  49. return self.area
  50. class Single(BaseCor):
  51. """
  52. 这个类使用于单行文本识别(所识别的ROI不会动)
  53. """
  54. def after_process(self, result):
  55. return result
  56. def ocr_single(self, image) -> str:
  57. """
  58. 检测某个固定位置的roi的文本。可以是横方向也可以是竖方向
  59. :param image:
  60. :return:
  61. """
  62. if self.roi:
  63. result = self.ocr_single_line(image)
  64. if result != "":
  65. return result
  66. # 如果没有识别到,这个时候考虑到可能是竖方向的文本, 使用detect_and_ocr来进行识别
  67. logger.info(f"OCR {self.name}: No text detected in ROI, try to detect vertically")
  68. result = self.detect_and_ocr(image)
  69. if not result:
  70. logger.info(f"OCR {self.name}: No text detected in ROI")
  71. return ""
  72. if result[0].ocr_text != "" and result[0].score > self.score:
  73. return result[0].ocr_text
  74. # 如果还是没有识别到。那可能就是真的没有识别到了
  75. return ""
  76. else:
  77. raise ScriptError("Roi is empty")
  78. class Digit(Single):
  79. def after_process(self, result):
  80. result = super().after_process(result)
  81. result = result.replace('I', '1').replace('D', '0').replace('S', '5')
  82. result = result.replace('B', '8')
  83. prev = result
  84. result = int(result) if result else 0
  85. if str(result) != prev:
  86. logger.warning(f'OCR {self.name}: Result "{prev}" is revised to "{result}"')
  87. return result
  88. def ocr_digit(self, image) -> int:
  89. """
  90. 返回数字
  91. :param image:
  92. :return:
  93. """
  94. result = self.ocr_single(image)
  95. if result == "":
  96. return 0
  97. else:
  98. return int(result)
  99. class DigitCounter(Single):
  100. def after_process(self, result):
  101. result = super().after_process(result)
  102. result = result.replace('I', '1').replace('D', '0').replace('S', '5')
  103. result = result.replace('B', '8')
  104. return result
  105. @classmethod
  106. def ocr_str_digit_counter(cls, result: str) -> tuple[int, int, int]:
  107. result = re.search(r'(\d+)/(\d+)', result)
  108. if result:
  109. result = [int(s) for s in result.groups()]
  110. current, total = int(result[0]), int(result[1])
  111. current = min(current, total)
  112. return current, total - current, total
  113. else:
  114. logger.warning(f'Unexpected ocr result: {result}')
  115. return 0, 0, 0
  116. def ocr_digit_counter(self, image) -> tuple[int, int, int]:
  117. """
  118. 获取计数的结果
  119. :param image:
  120. :return: 例如 14/15,返回 (14, 1, 15) 。如果没有识别到,返回 (0, 0, 0)
  121. """
  122. result = self.ocr_single(image)
  123. if result == "":
  124. return 0, 0, 0
  125. return self.ocr_str_digit_counter(result)
  126. class Duration(Single):
  127. def after_process(self, result):
  128. result = super().after_process(result)
  129. result = result.replace('I', '1').replace('D', '0').replace('S', '5')
  130. result = result.replace('B', '8')
  131. return result
  132. @staticmethod
  133. def parse_time(string):
  134. """
  135. Args:
  136. string (str): `01:30:00`
  137. Returns:
  138. datetime.timedelta:
  139. """
  140. result = re.search(r'(\d{1,2}):?(\d{2}):?(\d{2})', string)
  141. if result:
  142. result = [int(s) for s in result.groups()]
  143. return timedelta(hours=result[0], minutes=result[1], seconds=result[2])
  144. else:
  145. logger.warning(f'Invalid duration: {string}')
  146. return timedelta(hours=0, minutes=0, seconds=0)
  147. def ocr_duration(self, image) -> timedelta:
  148. """
  149. :param image:
  150. :return:
  151. """
  152. result = self.ocr_single(image)
  153. if result == "":
  154. return timedelta(hours=0, minutes=0, seconds=0)
  155. return self.parse_time(result)
  156. if __name__ == '__main__':
  157. d = DigitCounter.ocr_str_digit_counter('14/77')
  158. print(d)
  159. u = Duration.parse_time('01:30:00')
  160. print(u)