OCR預處理：矯正圖片中的文本信息（opencv）...

黃爸爸好 2021-03-30

展開全文

1.首先導入工程所需要的三方包,，這里需要opencv,、numpy,、math

import cv2
import numpy as np
import math

2.讀取圖片

img = cv2.imread(file_path)

3.圖片去噪

img_c = cv2.fastNlMeansDenoisingColored(img, None, 10, 10, 7, 21)

4.處理成灰度圖

gray = cv2.cvtColor(img_c, cv2.COLOR_BGR2GRAY)

5.Sobel算子,，x方向求梯度,主要用于獲得數(shù)字圖像的一階梯度，常見的應用和物理意義是邊緣檢測,。（可以根據(jù)需求選擇算子）

 sobel = cv2.Sobel(gray, cv2.CV_8U, 1, 0, ksize=3)

6.二值化

ret, binary = cv2.threshold(sobel, 0, 255, cv2.THRESH_OTSU + cv2.THRESH_BINARY)

7.霍夫直線

    hufu = binary.astype(np.uint8)
    lines = cv2.HoughLinesP(hufu, 1, np.pi / 180, 30, minLineLength=40, maxLineGap=100)

    # 在圖像上展示霍夫直線描出的直線
    # for line in lines:
    #     cv2.line(img, (line[0][0], line[0][1]), (line[0][2], line[0][3]), (0, 0, 255), 2)

8.求出所有直線斜率,，求出眾數(shù)（考慮誤差），這一步會產(chǎn)生個bug,，因為這些直線是利用霍夫直線找到的,，所以說需要根據(jù)實際場景來設置霍夫直線的相關參數(shù)，不然會對圖片中的線性非常敏感,，比如條形碼

    k_dict = {}
    k = 0

    
    for line in lines:
        if line[0][2] - line[0][0] == 0:
            continue
        print(line[0][3],line[0][1],line[0][2],line[0][0])
        k = (line[0][3] - line[0][1]) / (line[0][2] - line[0][0])
        # α = atan(k) * 180 / PI
        k = math.atan(k) * 180 / np.pi
        if len(k_dict.keys()) == 0 :
            k_dict[k] = 1
        else:
            flag = False
            for item in k_dict.keys():
                if abs(item - k) < 2:
                    flag = True
                    k_dict[item] += 1
                    break
            if not flag:
                k_dict[k] = 1

    must_k_num = 0
    must_key = 0
    for item in k_dict.keys():
        if k_dict[item] > must_k_num:
            must_k_num = k_dict[item]
            must_key = item


    print(must_key)

9.旋轉(zhuǎn)圖像,，在旋轉(zhuǎn)圖像之前需要對圖片進行填充防止旋轉(zhuǎn)后邊角溢出（這一步可以根據(jù)角度和勾股定理來計算精準的填充大小）,，利用仿射變換來旋轉(zhuǎn)圖像

#旋轉(zhuǎn)圖像

    h, w = img.shape[:2]
    add_w = int((((w*w + h*h) ** 0.5) - w)/2)
    add_h = int((((w*w + h*h) ** 0.5) - h)/2)
    print(add_w,add_h)

    img = cv2.copyMakeBorder(img,add_h,add_h,add_w,add_w, cv2.BORDER_CONSTANT,value=[0,0,0])

    h, w = img.shape[:2]
    center = (w//2, h//2)


    M = cv2.getRotationMatrix2D(center, must_key, 1.0)
    rotated = cv2.warpAffine(img, M, (w, h), flags=cv2.INTER_CUBIC)

    cv2.imwrite(file_path,rotated)


    cv2.imshow("rotated", rotated)
    cv2.waitKey(0)

全部代碼在這?。。,。,。?！

import cv2
import numpy as np
import math

def rotate(file_path):
    img = cv2.imread(file_path)

    #去噪
    img_c = cv2.fastNlMeansDenoisingColored(img, None, 10, 10, 7, 21)

    # 灰度圖
    gray = cv2.cvtColor(img_c, cv2.COLOR_BGR2GRAY)
    # cv2.imshow("gray", gray)


    # Sobel算子,，x方向求梯度,主要用于獲得數(shù)字圖像的一階梯度,，常見的應用和物理意義是邊緣檢測,。
    sobel = cv2.Sobel(gray, cv2.CV_8U, 1, 0, ksize=3)

    # 二值化
    ret, binary = cv2.threshold(sobel, 0, 255, cv2.THRESH_OTSU + cv2.THRESH_BINARY)


    # 霍夫直線
    hufu = binary.astype(np.uint8)
    lines = cv2.HoughLinesP(hufu, 1, np.pi / 180, 30, minLineLength=40, maxLineGap=100)

    # for line in lines:
    #     cv2.line(img, (line[0][0], line[0][1]), (line[0][2], line[0][3]), (0, 0, 255), 2)


    k_dict = {}
    k = 0

    #求出所有直線斜率，求出眾數(shù)（考慮誤差）
    for line in lines:
        if line[0][2] - line[0][0] == 0:
            continue
        print(line[0][3],line[0][1],line[0][2],line[0][0])
        k = (line[0][3] - line[0][1]) / (line[0][2] - line[0][0])
        # α = atan(k) * 180 / PI
        k = math.atan(k) * 180 / np.pi
        if len(k_dict.keys()) == 0 :
            k_dict[k] = 1
        else:
            flag = False
            for item in k_dict.keys():
                if abs(item - k) < 2:
                    flag = True
                    k_dict[item] += 1
                    break
            if not flag:
                k_dict[k] = 1

    must_k_num = 0
    must_key = 0
    for item in k_dict.keys():
        if k_dict[item] > must_k_num:
            must_k_num = k_dict[item]
            must_key = item


    print(must_key)

    #旋轉(zhuǎn)圖像

    h, w = img.shape[:2]
    add_w = int((((w*w + h*h) ** 0.5) - w)/2)
    add_h = int((((w*w + h*h) ** 0.5) - h)/2)
    print(add_w,add_h)

    img = cv2.copyMakeBorder(img,add_h,add_h,add_w,add_w, cv2.BORDER_CONSTANT,value=[0,0,0])

    h, w = img.shape[:2]
    center = (w//2, h//2)


    M = cv2.getRotationMatrix2D(center, must_key, 1.0)
    rotated = cv2.warpAffine(img, M, (w, h), flags=cv2.INTER_CUBIC)

    cv2.imwrite(file_path,rotated)


    cv2.imshow("rotated", rotated)
    cv2.waitKey(0)