python利用opencv实现印章的提取

朱雀 2020-10-18 06:44 3113阅读 5赞

抽空写了下，怕找不到了，做个备份，直接上代码，嗯，自己能看懂就喜欢这种感觉：

#coding:utf-8
    import cv2
    import numpy as np
    
    #加载图片
    image = cv2.imread('img\\111.jpg')
    #统一处理图片大小
    img_w = 650 if image.shape[1] > 600 else 400
    image = cv2.resize(image, (img_w,int(img_w*image.shape[0]/image.shape[1])), interpolation=cv2.IMREAD_COLOR)
    impng=cv2.cvtColor(image.copy(), cv2.COLOR_RGB2RGBA)
    # image = cv2.cvtColor(image, cv2.COLOR_RGBA2BGR)
    #image = cv2.imread('img\\public-2004-16-a7912dbd-72b0-4d96-9d65-70abc21940d5.png')
    hue_image = cv2.cvtColor(image, cv2.COLOR_BGR2HSV)
    low_range = np.array([130, 43, 46])
    high_range = np.array([180, 255, 255])
    th = cv2.inRange(hue_image, low_range, high_range)
    element = cv2.getStructuringElement(cv2.MORPH_RECT, (1, 1))
    th = cv2.dilate(th, element)
    index1 = th == 255
    img1 = np.zeros(impng.shape, np.uint8)
    img1[:,:,:] = (255, 255, 255,0)
    img1[index1] = impng[index1]  # (0,0,255)
    
    # img1 = cv2.cvtColor(img1, cv2.COLOR_BGR2RGB)
    # 创建四通道图片
    #img1=topng(img1)
    #img1 = cv2.cvtColor(img1, cv2.COLOR_RGB2RGBA)
    cv2.imwrite('m1.png', img1)
    cv2.imshow('1', img1)
    #cv2.waitKey()
    # cv2.imwrite("img1.png", img1)
    low_range = np.array([0, 43, 46])
    high_range = np.array([9, 255, 255])
    th = cv2.inRange(hue_image, low_range, high_range)
    element = cv2.getStructuringElement(cv2.MORPH_RECT, (1, 1))
    th = cv2.dilate(th, element)
    index1 = th == 255
    img2 = np.zeros(impng.shape, np.uint8)
    img2[:,:,:] = (255, 255, 255,0)
    img2[index1] = impng[index1]
    cv2.imshow('2', img2)
    cv2.imwrite('m2.png', img2)
    
    imgreal=cv2.add(img2,img1)
    cv2.imwrite('m222.png', imgreal)
    #cv2.waitKey()
    #img31 = cv2.cvtColor(img31, cv2.COLOR_BGR2RGB)
    white_px = np.asarray([255, 255, 255,255])
    #black_px = np.asarray([0, 0, 0, 255])
    (row, col, _) = imgreal.shape
    for r in range(row):
        for c in range(col):
            px = imgreal[r][c]
            if all(px == white_px):
                imgreal[r][c] = impng[r][c]
    
    #扩充图片防止截取部分
    img4png=cv2.copyMakeBorder(imgreal, 50, 50, 50, 50,cv2.BORDER_CONSTANT, value=[255, 255, 255,0])
    img2gray = cv2.cvtColor(img4png, cv2.COLOR_RGBA2GRAY)
    retval, grayfirst = cv2.threshold(img2gray, 254, 255, cv2.THRESH_BINARY_INV)
    
    # 再次膨胀，轮廓查找
    # img6 = cv2.GaussianBlur(grayfirst, (25, 5), 0, 0)
    element = cv2.getStructuringElement(cv2.MORPH_RECT, (22, 22))
    img6 = cv2.dilate(grayfirst, element)
    #cv2.imshow('img4', img6)
    
    c_canny_img = cv2.Canny(img6, 10, 10)
    
    contours, hierarchy = cv2.findContours(c_canny_img, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
    areas = []
    for i, cnt in enumerate(contours):
        rect = cv2.minAreaRect(cnt)
        x, y, w, h = cv2.boundingRect(cnt)
        area = w * h
        ars = [area, i]
        areas.append(ars)
    areas = sorted(areas, reverse=True)
    print(areas)
    maxares = areas[:1]
    
    x, y, w, h = cv2.boundingRect(contours[maxares[0][1]])
    temp = img4png[y:(y + h), x:(x + w)]
    #高小于宽
    print(temp.shape)
    if temp.shape[0]<temp.shape[1]:
        #temp = cv2.resize(temp, (150, temp.shape[0]))
        zh=int((temp.shape[1]-temp.shape[0])/2)
        temp = cv2.copyMakeBorder(temp, zh, zh, 0, 0, cv2.BORDER_CONSTANT, value=[255, 255, 255,0])
    else:
        #temp = cv2.resize(temp, (temp.shape[1], 150))
        zh = int((temp.shape[0] - temp.shape[1]) / 2)
        temp = cv2.copyMakeBorder(temp, 0,0,zh, zh, cv2.BORDER_CONSTANT, value=[255, 255, 255,0])
    dst = cv2.resize(temp, (150, 150))
    
    cv2.imshow('3', dst)
    cv2.imwrite('m0.png', dst)
    cv2.waitKey()

看看效果：

![2020060820313558.jpg][]    ![20200608203221184.png][]

[2020060820313558.jpg]: /images/1603003438758.jpg
[20200608203221184.png]: /images/1603003390981.png