hr-ocr-cet/core/direction.py

import re
from dataclasses import dataclass
from enum import Enum
from typing import Tuple, List

import cv2
import numpy as np
from paddleocr import PaddleOCR

from core.line_parser import LineParser


class Direction(Enum):
    TOP = 0
    RIGHT = 1
    BOTTOM = 2
    LEFT = 3


# 父类
class OcrAnchor(object):
    # 输入识别anchor的名字, 如身份证号
    def __init__(self, name: str, d: List[Direction]):
        self.name = name
        # anchor位置
        self.direction = d

        def t_func(anchor, c, is_horizontal):
            if is_horizontal:
                return 0 if anchor[1] < c[1] else 2
            else:
                return 1 if anchor[0] > c[0] else 3

        def l_func(anchor, c, is_horizontal):
            if is_horizontal:
                return 0 if anchor[0] < c[0] else 2
            else:
                return 1 if anchor[1] < c[1] else 3

        def b_func(anchor, c, is_horizontal):
            if is_horizontal:
                return 0 if anchor[1] > c[1] else 2
            else:
                return 1 if anchor[0] < c[0] else 3

        def r_func(anchor, c, is_horizontal):
            if is_horizontal:
                return 0 if anchor[0] > c[0] else 2
            else:
                return 1 if anchor[1] > c[1] else 3

        self.direction_funcs = {
            Direction.TOP: t_func,
            Direction.BOTTOM: b_func,
            Direction.LEFT: l_func,
            Direction.RIGHT: r_func,
        }

    # 获取中心区域坐标 -> (x, y)
    def get_rec_area(self, res) -> Tuple[float, float]:
        """获得整张身份证的识别区域, 返回识别区域的中心点"""
        boxes = []
        for row in res:
            for r in row:
                boxes.extend(r.box)
        boxes = np.stack(boxes)
        l, t = np.min(boxes, 0)
        r, b = np.max(boxes, 0)
        # 识别区域的box
        # big_box = [[l, t], [r, t], [r, b], [l, b]]
        # w, h = (r - l, b - t)
        return (l + r) / 2, (t + b) / 2

    # 判断是否是 锚点
    def is_anchor(self, txt, box) -> bool:
        pass

    # 找 锚点 -> 锚点坐标
    def find_anchor(self, res) -> Tuple[bool, float, float]:
        """
        寻找锚点 中心点坐标
        """
        for row in res:
            for r in row:
                txt = r.txt.replace('-', '').replace(' ', '')
                box = r.box
                if self.is_anchor(txt, box):
                    l, t = np.min(box, 0)
                    r, b = np.max(box, 0)
                    return True, (l + r) / 2, (t + b) / 2
        return False, 0., 0.

    # 定位 锚点 -> 角度
    # -> 锚点(x, y)  pic(x, y) is_horizontal
    def locate_anchor(self, res, is_horizontal) -> int:
        found, id_cx, id_cy = self.find_anchor(res)

        # 如果识别不到身份证号
        if not found: raise Exception(f'识别不到anchor{self.name}')
        cx, cy = self.get_rec_area(res)
        # print(f'id_cx: {id_cx}, id_cy: {id_cy}')
        # print(f'cx: {cx}, cy: {cy}')
        pre = None
        for d in self.direction:
            f = self.direction_funcs.get(d, None)
            angle = f((id_cx, id_cy), (cx, cy), is_horizontal)
            if pre is None:
                pre = angle
            else:
                if angle != pre:
                    raise Exception('angle is not compatiable')
        return pre

        # if is_horizontal:
        #     # 如果是水平的，身份证号的位置在相对识别区域的下方，方向则为0度，否则是180度
        #     return 0 if id_cy > cy else 2
        # else:
        #     # 如果是竖直的，身份证号的相对位置如果在左边，方向为90度，否则270度
        #     return 1 if id_cx < cx else 3


# 子类1 人像面
class CETAnchor(OcrAnchor):
    def __init__(self, name: str, d: List[Direction]):
        super(CETAnchor, self).__init__(name, d)

    def is_anchor(self, txt, box) -> bool:
        txts = re.findall('全国大学英语', txt)
        if len(txts) > 0:
            return True
        return False

    def locate_anchor(self, res, is_horizontal) -> int:
        return super(CETAnchor, self).locate_anchor(res, is_horizontal)


# 子类2 国徽面
class TEMAnchor(OcrAnchor):
    def __init__(self, name: str, d: List[Direction]):
        super(TEMAnchor, self).__init__(name, d)

    def is_anchor(self, txt, box) -> bool:
        txts = re.findall('证书编号', txt)
        if len(txts) > 0:
            return True
        return False

    def locate_anchor(self, res, is_horizontal) -> int:
        return super(TEMAnchor, self).locate_anchor(res, is_horizontal)


# 调用以上 🔧工具
# <- ocr_生数据
# == ocr_熟数据(行处理后)
# -> 角度0/1/2/3
def detect_angle(result, ocr_anchor: OcrAnchor):
    filters = [lambda x: x.is_slope, lambda x: x.txt.replace(' ', '').encode('utf-8').isalpha()]
    lp = LineParser(result, filters)
    res = lp.parse()
    print('------ angle ocr -------')
    print(res)
    print('------ angle ocr -------')
    is_horizontal = lp.is_horizontal
    return ocr_anchor.locate_anchor(res, is_horizontal)


@dataclass
class AngleDetector(object):
    """
    角度检测器
    """
    ocr: PaddleOCR

    # 角度检测器
    # <- img(cv2格式)  img_type
    # == result <- img(cv2)
    # -> angle       result(ocr生)
    def detect_angle(self, img):
        # image_type = int(image_type)
        # result = self.ocr.ocr(img, cls=True)

        image_type, result = self.detect_img(img)

        ocr_anchor = CETAnchor('CET', [Direction.TOP]) if image_type == 0 else TEMAnchor('TEM', [
            Direction.BOTTOM])

        try:
            angle = detect_angle(result, ocr_anchor)
            return angle, result, image_type

        except Exception as e:
            print(e)
            # 如果第一次识别不到，旋转90度再识别
            img = cv2.rotate(img, cv2.ROTATE_90_CLOCKWISE)
            result = self.ocr.ocr(img, cls=True)
            angle = detect_angle(result, ocr_anchor)
            # 旋转90度之后要重新计算角度
            return (angle - 1 + 4) % 4, result, image_type

    def detect_img(self, img):
        result = self.ocr.ocr(img, cls=True)
        for res in result:
            if "报告单" in res[1][0]:
                return 0, result
        raise Exception("不支持专四专八")
        # return 1, result