14.基于opencv实现文字识别

本是古典何须时尚 2024-04-07 12:24 74阅读 0赞

from typing import Tuple
    import cv2
    import numpy as np
    
    def grayScale(img : np.ndarray) -> np.ndarray:
           return cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
    
    def blur(img : np.ndarray, k: int = 3, std : int = 0) -> np.ndarray:
           return cv2.GaussianBlur(img, (k,k), std)
    
    def binary(img : np.ndarray) -> np.ndarray:
           return cv2.Canny(img, 100, 255)
    
    def retBiggestContour(img : np.ndarray) -> Tuple[np.ndarray, Tuple[int, int]]:
        contours, _ = cv2.findContours(img, cv2.RETR_LIST, cv2.CHAIN_APPROX_SIMPLE)
        cnt = sorted(contours, key = cv2.contourArea)[-1]
        _, (w,h), angle = cv2.minAreaRect(cnt)
        w,h = int(w), int(h)
        if angle > 45 or angle < -45:
            w,h = h,w
        return cnt, (w,h)
    
    def extract(img : np.ndarray, 
                cnt : np.ndarray, 
                w : int, 
                h : int
                ) -> np.ndarray:
        cnt = cnt.reshape(cnt.shape[0], cnt.shape[-1])
        # the corners have extreme x, y coordinates
        s1 = sorted(cnt, key = lambda x : (x[0], x[1]))
        s2 = sorted(cnt, key = lambda x : (x[1], x[0]))
        corner1, corner3 = s1[0], s1[-1]
        corner2, corner4 = s2[0], s2[-1]
    
        corners = np.array([corner1, corner2, corner3, corner4])
        target_corners = np.array([(0,0), (w,0), (w,h), (0,h)])
    
        H, _ = cv2.findHomography(corners, target_corners, params = None)
    
        transformed_image = cv2.warpPerspective(
        img, H, (img.shape[1], img.shape[0]))
    
        transformed_image = transformed_image[:h, :w]
    
        return transformed_image
    
    def transform(img : np.ndarray) -> np.ndarray:
        T = cv2.GaussianBlur(img, (11,11),0)-10
        return (img > T).astype(np.uint8) * 255
    
    if __name__ == "__main__":
        img = cv2.imread("G:/py/i/wenzi.jpg")
        gray = grayScale(img)
        blurred = blur(gray)
        edged = binary(blurred)
        cnt, (w,h) = retBiggestContour(edged)
        img = extract(gray, cnt, w, h)
        img = transform(img)
        cv2.namedWindow('img',cv2.WINDOW_NORMAL)
        cv2.imshow('img',img)
        k = cv2.waitKey(0)&0xFF
        if k == 27:
            cv2.destroyAllWindows()
        elif k == ord('q'): 
            cv2.destroyAllWindows()
        #cv2.imwrite("G:/transformed.jpg", img)