python-opencv surface defect detection (pattern recognition)

For many industrial inspections, especially the screening of some sophisticated devices, it is often at the pixel level, which is very accurate.

The main idea

• Convert image to binary image
• Erosion/dilation of the image
• Contour detection is in progress
• Filter the contours that match the target size (excluding contours with small errors)
• In the expansion process, the outline information is drawn

import cv2
import os
import numpy as np
import time


t1 = time.time()
img = cv2.imread('./label/28901647.jpg', 0)
img_copy = cv2.imread('./label/28901647.jpg', 0)
mask = np.zeros_like(img)
print(np.shape(img))
# 先利用二值化去除图片噪声
ret, img = cv2.threshold(img, 80, 255, cv2.THRESH_BINARY)




es = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (4, 2))
img = cv2.dilate(img, es, iterations=1)  # 形态学膨胀


kernel = np.ones(shape=[5,5],dtype=np.uint8)
img = cv2.erode(img,kernel=kernel)  # 腐蚀操作

cv2.imshow('aa',img)
cv2.waitKey(0)


contours, _ = cv2.findContours(img, cv2.RETR_TREE, cv2.CHAIN_APPROX_SIMPLE)


n = len(contours)  # 轮廓的个数
cv_contours = []
for contour in contours:
    area = cv2.contourArea(contour)

    if area <= 500:# 筛选面积大于500的，小于500的全部变为255，
        cv_contours.append(contour)
        # 方式一
        # x, y, w, h = cv2.boundingRect(contour) # 这个函数可以获得一个图像的最小矩形边框一些信息，参数img是一个二值图像，它可以返回四个参数，左上角坐标，矩形的宽高 (轮廓集合  contour)
        # img[y:y + h, x:x + w] = 255
        
    else:

        cv2.drawContours(img_copy, [contour], -1, (0, 0, 255), 0) # 多边形轮廓绘制

        print('area:', area)
        continue
# 方式二
cv2.fillPoly(img, cv_contours, (255, 255, 255)) # 多个多边形填充

t2 = time.time()
print('时间：',t2-t1)
cv2.imwrite('./output/28901647.jpg', img)

1. The found contour information (defects)

2. After corrosion and expansion, screened out较大缺陷

3. Screening through different degrees of expansion corrosion and defect area

Hope this works for you!
Thank you for your likes and comments!