如何提高对表面缺陷的检测？

Question

首先，这是我尝试检测拉丝铝表面上的缺陷（平行线）的原始图像。

这是我采取的步骤：

1. 高斯模糊
2. 放大图像
3. 将图像转换为灰度
4. 变形关闭操作
5. 再次扩张
6. 图像差异
7. Canny 边缘检测
8. 寻找轮廓
9. 在等高线周围画一条绿线

这是我的代码：

import numpy as np
import cv2
from matplotlib import pyplot as plt
import imutils
path = ''
path_output = ''

img_bgr = cv2.imread(path)
plt.imshow(img_bgr)

# bgr to rgb
img_rgb = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2RGB)
plt.imshow(img_rgb)

# Converting to grayscale
img_just_gray = cv2.cvtColor(img_rgb, cv2.COLOR_BGR2GRAY)

# Displaying the grayscale image
plt.imshow(img_just_gray, cmap='gray')

# Gaussian Blur
ksize_w = 13
ksize_h = 13

img_first_gb = cv2.GaussianBlur(img_rgb, (ksize_w,ksize_h), 0, 0, cv2.BORDER_REPLICATE);
plt.imshow(img_first_gb)

# Dilate the image

dilated_img = cv2.dilate(img_first_gb, np.ones((11,11), np.uint8))
plt.imshow(dilated_img)

# Converting to grayscale
img_gray_operated = cv2.cvtColor(dilated_img, cv2.COLOR_BGR2GRAY)

# Displaying the grayscale image
plt.imshow(img_gray_operated, cmap='gray')

# closing:
kernel_closing = np.ones((7,7),np.uint8)
img_closing = cv2.morphologyEx(img_gray_operated, cv2.MORPH_CLOSE, kernel_closing)
plt.imshow(img_closing, cmap='gray')

# dilation:
# add pixels to the boundaries of objects in an image
kernel_dilation = np.ones((3,3),np.uint8)
img_dilation2 = cv2.dilate(img_closing, kernel_dilation, iterations = 1)
plt.imshow(img_dilation2, cmap='gray')

diff_img = 255 - cv2.absdiff(img_just_gray, img_dilation2)
plt.imshow(diff_img, cmap='gray')

# canny
edgesToFindImage = img_dilation2

v = np.median(img_just_gray)
#print(v)
sigma = 0.33
lower_thresh = int(max(0,(1.0-sigma)*v))
higher_thresh = int(min(255,(1.0+sigma)*v))

img_edges =  cv2.Canny(edgesToFindImage, lower_thresh, higher_thresh)
plt.imshow(img_edges, cmap='gray')

kernel_dilation2 = np.ones((2,2),np.uint8)
img_dilation2 = cv2.dilate(img_edges, kernel_dilation, iterations = 2)
plt.imshow(img_dilation2, cmap='gray')

# find contours
contoursToFindImage = img_dilation2

(_, cnts, _) = cv2.findContours(contoursToFindImage.copy(), cv2.RETR_EXTERNAL,
        cv2.CHAIN_APPROX_SIMPLE)
print(type(cnts))
print(len(cnts))

# -1 for all
cntsWhichOne = -1

# -1 for infill
# >0 for edge thickness
cntsInfillOrEdgeThickness = 3

img_drawing_contours_on_rgb_image = cv2.drawContours(img_rgb.copy(), cnts, cntsWhichOne, (0, 255, 0), cntsInfillOrEdgeThickness)
plt.imshow(img_drawing_contours_on_rgb_image)

这就是结果。

如何改进这种检测？有没有更有效的检测线的方法？

Answer 1

这是 Python OpenCV 中的一种方法。你已经接近了，你应该使用自适应阈值，形态学来清理小区域并跳过精明的边缘。

输入：

import cv2
import numpy as np

# load image
img = cv2.imread('scratches.jpg')

# convert to grayscale
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)

# adaptive threshold 
thresh = cv2.adaptiveThreshold(gray, 255, cv2.ADAPTIVE_THRESH_MEAN_C, cv2.THRESH_BINARY, 11, -35)

# apply morphology
kernel = np.ones((3,30),np.uint8)
morph = cv2.morphologyEx(thresh, cv2.MORPH_CLOSE, kernel)
kernel = np.ones((3,35),np.uint8)
morph = cv2.morphologyEx(morph, cv2.MORPH_OPEN, kernel)

# get hough line segments
threshold = 25
minLineLength = 10
maxLineGap = 20
lines = cv2.HoughLinesP(morph, 1, 30*np.pi/360, threshold, minLineLength, maxLineGap)

# draw lines
linear1 = np.zeros_like(thresh)
linear2 = img.copy()
for [line] in lines:
    x1 = line[0]
    y1 = line[1]
    x2 = line[2]
    y2 = line[3]
    cv2.line(linear1, (x1,y1), (x2,y2), 255, 1)
    cv2.line(linear2, (x1,y1), (x2,y2), (0,0,255), 1)

print('number of lines:',len(lines))

# save resulting masked image
cv2.imwrite('scratches_thresh.jpg', thresh)
cv2.imwrite('scratches_morph.jpg', morph)
cv2.imwrite('scratches_lines1.jpg', linear1)
cv2.imwrite('scratches_lines2.jpg', linear2)

# display result
cv2.imshow("thresh", thresh)
cv2.imshow("morph", morph)
cv2.imshow("lines1", linear1)
cv2.imshow("lines2", linear2)
cv2.waitKey(0)
cv2.destroyAllWindows()

阈值图像：

形态清洁图像：

原图上的线条：

黑色背景上的线条：