| 123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291 |
- # This Python file uses the following encoding: utf-8
- # @author runhey
- # github https://github.com/runhey
- import time
- import cv2
- import numpy as np
- from ppocronnx.predict_system import BoxedResult
- from enum import Enum
- from module.base.decorator import cached_property
- from module.base.utils import area_pad, crop, float2str
- from module.ocr.ppocr import TextSystem
- from module.ocr.models import OCR_MODEL
- from module.exception import ScriptError
- from module.logger import logger
- def enlarge_canvas(image):
- """
- copy from https://github.com/LmeSzinc/StarRailCopilot
- Enlarge image into a square fill with black background. In the structure of PaddleOCR,
- image with w:h=1:1 is the best while 3:1 rectangles takes three times as long.
- Also enlarge into the integer multiple of 32 cause PaddleOCR will downscale images to 1/32.
- """
- height, width = image.shape[:2]
- length = int(max(width, height) // 32 * 32 + 32)
- border = (0, length - height, 0, length - width)
- if sum(border) > 0:
- image = cv2.copyMakeBorder(image, *border, borderType=cv2.BORDER_CONSTANT, value=(0, 0, 0))
- return image
- class OcrMode(Enum):
- FULL = 1 # str: "Full"
- SINGLE = 2 # str: "Single"
- DIGIT = 3 # str: "Digit"
- DIGITCOUNTER = 4 # str: "DigitCounter"
- DURATION = 5 # str: "Duration"
- class OcrMethod(Enum):
- DEFAULT = 1 # str: "Default"
- class BaseCor:
- lang: str = "ch"
- score: float = 0.6 # 阈值默认为0.5
- name: str = "ocr"
- mode: OcrMode = OcrMode.FULL
- method: OcrMethod = OcrMethod.DEFAULT # 占位符
- roi: list = [] # [x, y, width, height]
- area: list = [] # [x, y, width, height]
- keyword: str = "" # 默认为空
- def __init__(self,
- name: str,
- mode: str,
- method: str,
- roi: tuple,
- area: tuple,
- keyword: str) -> None:
- """
- :param name:
- :param mode:
- :param method:
- :param roi:
- :param area:
- :param keyword:
- """
- self.name = name.upper()
- if isinstance(mode, str):
- self.mode = OcrMode[mode.upper()]
- elif isinstance(mode, OcrMode):
- self.mode = mode
- if isinstance(method, str):
- self.method = OcrMethod[method.upper()]
- elif isinstance(method, OcrMethod):
- self.method = method
- self.roi: list = list(roi)
- self.area: list = list(area)
- self.keyword = keyword
- @cached_property
- def model(self) -> TextSystem:
- return OCR_MODEL.__getattribute__(self.lang)
- def pre_process(self, image):
- """
- 重写
- :param image:
- :return:
- """
- return image
- def after_process(self, result):
- """
- 重写
- :param result:
- :return:
- """
- return result
- @classmethod
- def crop(cls, image: np.array, roi: tuple) -> np.array:
- """
- 截取图片
- :param roi:
- :param image:
- :return:
- """
- x, y, w, h = roi
- return image[y:y + h, x:x + w]
- def ocr_item(self, image):
- """
- 这个函数区别于ocr_single_line,这个函数不对图片进行裁剪,是什么就是什么的
- :param image:
- :return:
- """
- # pre process
- start_time = time.time()
- image = self.pre_process(image)
- # ocr
- result, score = self.model.ocr_single_line(image)
- if score < self.score:
- result = ""
- # after proces
- result = self.after_process(result)
- # logger.info("ocr result score: %s%s" % (result,score))
- logger.attr(name='%s %ss' % (self.name, float2str(time.time() - start_time)),
- text=f'[{result}]')
- return result
- def ocr_single_line(self, image):
- """
- 只支持横方向的单行ocr,不支持竖方向的单行ocr
- 注意:这里使用了预处理和后处理
- :param image:
- :return:
- """
- # pre process
- start_time = time.time()
- image = self.crop(image, self.roi)
- image = self.pre_process(image)
- # ocr
- result, score = self.model.ocr_single_line(image)
- if score < self.score:
- result = ""
- # after proces
- result = self.after_process(result)
- # logger.info("ocr result score: %s" % score)
- logger.attr(name='%s %ss' % (self.name, float2str(time.time() - start_time)),
- text=f'[{result}]')
- return result
- def detect_and_ocr(self, image) -> list[BoxedResult]:
- """
- 注意:这里使用了预处理和后处理
- :param image:
- :return:
- """
- # pre process
- start_time = time.time()
- image = self.crop(image, self.roi)
- image = self.pre_process(image)
- image = enlarge_canvas(image)
- # ocr
- boxed_results: list[BoxedResult] = self.model.detect_and_ocr(image)
- results = []
- # after proces
- for result in boxed_results:
- # logger.info("ocr result score: %s" % result.score)
- if result.score < self.score:
- continue
- result.ocr_text = self.after_process(result.ocr_text)
- results.append(result)
- logger.attr(name='%s %ss' % (self.name, float2str(time.time() - start_time)),
- text=str([result.ocr_text for result in results]))
- return results
- def match(self, result: str, included: bool=False) -> bool:
- """
- 使用ocr获取结果后和keyword进行匹配
- :param result:
- :param included: ocr结果和keyword是否包含关系, 要么是包含关系,要么是相等关系
- :return:
- """
- if included:
- return self.keyword in result
- else:
- return self.keyword == result
- def filter(self, boxed_results: list[BoxedResult], keyword: str=None) -> list or None:
- """
- 使用ocr获取结果后和keyword进行匹配. 返回匹配的index list
- :param keyword: 如果不指定默认适用对象的keyword
- :param boxed_results:
- :return:
- """
- # 首先先将所有的ocr的str顺序拼接起来, 然后再进行匹配
- result = None
- strings = [boxed_result.ocr_text for boxed_result in boxed_results]
- concatenated_string = "".join(strings)
- if keyword is None:
- keyword = self.keyword
- if keyword in concatenated_string:
- result = [index for index, word in enumerate(strings) if keyword in word]
- else:
- result = None
- if result is not None:
- # logger.info("Filter result: %s" % result)
- return result
- # 如果适用顺序拼接还是没有匹配到,那可能是竖排的,使用单个字节的keyword进行匹配
- indices = []
- # 对于keyword中的每一个字符,都要在strings中进行匹配
- # 如果这个字符在strings中的某一个string中,那么就记录这个string的index
- max_index = len(strings) - 1
- for index, char in enumerate(keyword):
- for i, string in enumerate(strings):
- if char not in string:
- continue
- if i <= max_index:
- indices.append(i)
- break
- if indices:
- # 剔除掉重复的index
- indices = list(set(indices))
- return indices
- else:
- return None
- def detect_text(self, image) -> str:
- """
- 识别图片中的文字, 会按照顺序拼接起来
- :param image:
- :return:
- """
- # pre process
- start_time = time.time()
- image = self.crop(image, self.roi)
- image = self.pre_process(image)
- image = enlarge_canvas(image)
- # ocr
- boxed_results: list[BoxedResult] = self.model.detect_and_ocr(image)
- results = ''
- # after proces
- for result in boxed_results:
- # logger.info("ocr result score: %s" % result.score)
- if result.score < self.score:
- continue
- results += result.ocr_text
- # logger.info("ocr result score: %s" % score)
- logger.attr(name='%s %ss' % (self.name, float2str(time.time() - start_time)),
- text=f'[{results}]')
- return results
- # def test():
- # # strings = ["探", "索"]
- # # keyword = "探索"
- # # strings是截图ocr的结果,keyword是要匹配的关键字
- # strings = ['123', '456', '789', '101112', '131415', '456']
- # keyword = '123456'
- #
- # indices = []
- # # 对于keyword中的每一个字符,都要在strings中进行匹配
- # # 如果这个字符在strings中的某一个string中,那么就记录这个string的index
- # max_index = len(strings) - 1
- # for index, char in enumerate(keyword):
- # for i, string in enumerate(strings):
- # if char not in string:
- # continue
- # if i <= max_index:
- # indices.append(i)
- # break
- # if indices:
- # # 剔除掉重复的index
- # indices = list(set(indices))
- # return indices
- # else:
- # return None
- # print(test())
|