hr-ocr-regbook/core/ocr.py

from dataclasses import dataclass

import cv2

from core.line_parser import LineParser
from core.parser import *
from core.direction import *
import numpy as np
from paddleocr import PaddleOCR


# <- 传入pic pic_type
# 1. 旋转pic  (to 正向)
# 2. 重写识别pic  (get res)
# 3. 行处理res  (get res)
# 4. 对res字段逻辑识别  (get dict)
# -> dict

@dataclass
class IdCardOcr:
    ocr: PaddleOCR
    # 角度探测器
    angle_detector: AngleDetector

    # master
    def predict(self, image: np.ndarray, image_type: str):
        img_type = int(image_type)

        image, angle, result = self._rotate_img(image, img_type)
        print(f'---------- detect angle: {angle} 图片角度 ----------')
        if angle != 0:
            _, _, result = self._ocr(image)

        return self._post_process(result, angle, image_type)

    # 检测角度
    def _rotate_img(self, image, image_type) -> (np.ndarray, int):
        angle, result = self.angle_detector.detect_angle(image, image_type)
        if angle == 1:
            image = cv2.rotate(image, cv2.ROTATE_90_COUNTERCLOCKWISE)
        if angle == 2:
            image = cv2.rotate(image, cv2.ROTATE_180)
        if angle == 3:
            image = cv2.rotate(image, cv2.ROTATE_90_CLOCKWISE)
        # print("图片的角度：",angle)
        return image, angle, result

    def _ocr(self, image):
        # 获取模型检测结果
        result = self.ocr.ocr(image, cls=True)
        print("------------------")
        print(result)
        if not result:
            raise Exception('无法识别')
        confs = [line[1][1] for line in result]

        # 将检测到的文字放到一个列表中
        txts = [line[1][0] for line in result]
        print("......................................")
        print(txts)
        print("......................................")
        return txts, confs, result

    def _post_process(self, result, angle: int, image_type: str):
        # 行处理

        line_parser = LineParser(result)
        line_result = line_parser.parse()
        conf = line_parser.confidence  # 平均conf

        # 0:常驻人口面
        # 1:户口本首页

        if int(image_type) == 0:
            pass
            parser = PeopleRegBookParser(line_result)
        elif int(image_type) == 1:
            pass
            parser = FrontRegBookParser(line_result)
        else:
            raise Exception('未传入 image_type')

        # 字段逻辑处理后对res(dict)
        ocr_res = parser.parse()

        res = {
            "confidence": conf,
            "img_type": str(image_type),
            "orientation": angle,  # 原angle是逆时针，转成顺时针
            **ocr_res
        }

        return res