找异物--求助各位大佬给点思路

如图需要找出产品中的异物

圈中的为异物

这个是原图

这个是无异物的图像

异物图像2

异物图像3

异物图像4

对应Halcon代码如下：

1. FilePaths := ['i (0).png', 'i (1).png', 'i (2).png', 'i (3).png', 'i (4).png', 'i (5).png']
   
2. 
   
3. for Index := 0 to 5 by 1
   
4.    
   
5.     tuple_select (FilePaths, Index, FilePath)
   
6.     read_image (Image, FilePath)
   
7.         
   
8.     scale_image_range(Image, ImageScaled, 20000, 60000)
   
9.     convert_image_type(ImageScaled, Image8u, 'byte')
   
10.    
    
11.     gen_disc_se(SE, 'byte', 31, 31, 1)
    
12.     gray_bothat (Image8u, SE, ImageBotHat)
    
13.    
    
14.     threshold(ImageBotHat, SmallRegions, 87, 255)
    
15.     connection(SmallRegions, ConnectedSmallRegions)
    
16.     select_shape (ConnectedSmallRegions, SelectedSmallRegions1, ['area'], ['and'], [5], [200])
    
17.    
    
18.     select_gray(SelectedSmallRegions1, Image8u, SelectedSmallRegions2, 'mean', 'and', 0, 30)
    
19.    
    
20.     threshold(Image8u, BigRegions, 0, 25)
    
21.     connection(BigRegions, ConnectedBigRegions)
    
22.     select_shape(ConnectedBigRegions, SelectedBigRegions, 'area', 'and', 100, 99999)
    
23.    
    
24.     concat_obj(SelectedBigRegions, SelectedSmallRegions2, ResultRegions)
    
25.    
    
26.     threshold(ImageBotHat, BoundaryDefectRegions, 110, 255)
    
27.    
    
28.     smooth_image(Image8u, ImageSmooth,'gauss', 30)
    
29.     threshold(ImageSmooth, BoundaryRegions, 90, 255)
    
30.     intersection(BoundaryRegions, BoundaryDefectRegions, BoundaryDefectRegions)
    
31.     connection(BoundaryDefectRegions, ConnectedBoundaryDefectRegions)
    
32.     select_shape(ConnectedBoundaryDefectRegions, SelectedBoundaryDefectRegions, 'area', 'and', 10, 99999)
    
33.     select_gray(SelectedBoundaryDefectRegions, Image8u, SelectedBoundaryDefectRegions2, 'mean', 'and', 0, 60)
    
34.    
    
35.     concat_obj(SelectedBoundaryDefectRegions2, ResultRegions, ResultRegions)
    
36.    
    
37.     gen_disc_se(SE, 'byte', 7, 7, 1)
    
38.     dilation1 (ResultRegions, SE, RegionDilation, 3)
    
39.    
    
40.     overpaint_region(Image8u, RegionDilation, 255, 'margin')
    
41.    
    
42.     tuple_str_replace (FilePath, '.png', '_result.png', ResultFilePath)
    
43.     write_image(Image8u, 'png', 0, ResultFilePath)
    
44. endfor
    
45. 
    
46. 
    

复制代码

OpenCV代码如下：

1. import cv2
   
2. import numpy as np
   
3. 
   
4. file_paths = ['i (1).png', 'i (2).png', 'i (3).png', 'i (4).png', 'i (5).png']
   
5. 
   
6. for index, file_path in enumerate(file_paths):
   
7.     # 读取图像（支持 8/16 位）
   
8.     image = cv2.imread(file_path, cv2.IMREAD_UNCHANGED)
   
9.    
   
10.     # 确保图像为 16 位
    
11.     if image.dtype == np.uint8:
    
12.         image = image.astype(np.uint16)
    
13. 
    
14.     image[image < 20000] = 20000
    
15.     image[image > 60000] = 60000
    
16.    
    
17.     image_8u = ((image - 20000) / (60000 - 20000) * 255).astype(np.uint8)
    
18.    
    
19.     kernel = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (31, 31))
    
20.     image_bot_hat = cv2.morphologyEx(image_8u, cv2.MORPH_BLACKHAT, kernel)
    
21. 
    
22.     _, small_regions = cv2.threshold(image_bot_hat, 87, 255, cv2.THRESH_BINARY)
    
23.     num_labels, labels = cv2.connectedComponents(small_regions, connectivity=8)
    
24. 
    
25.     selected_small_regions1 = np.zeros_like(labels)
    
26.     for i in range(1, num_labels):
    
27.         area = np.sum(labels == i)
    
28.         if 5 <= area <= 200:
    
29.             selected_small_regions1[labels == i] = 255
    
30.    
    
31.     selected_small_regions2 = np.zeros_like(labels)
    
32.     for i in range(1, num_labels):
    
33.         if np.any(selected_small_regions1 == 255):
    
34.             mean_val = np.mean(image_8u[labels == i])
    
35.             if 0 <= mean_val <= 30:
    
36.                 selected_small_regions2[labels == i] = 255
    
37.    
    
38.     _,  big_regions = cv2.threshold(image_8u, 25, 255, cv2.THRESH_BINARY_INV)
    
39.     num_labels_big, labels_big = cv2.connectedComponents(big_regions, connectivity=8)
    
40.    
    
41.     selected_big_regions = np.zeros_like(labels_big)
    
42.     for i in range(1, num_labels_big):
    
43.         area = np.sum(labels_big == i)
    
44.         if 100 <= area <= 99999:
    
45.             selected_big_regions[labels_big == i] = 255
    
46. 
    
47.     result_regions = np.bitwise_or(selected_small_regions2, selected_big_regions)
    
48.    
    
49.     kernel_dilate = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (7, 7))
    
50.     region_dilation = cv2.dilate(result_regions.astype(np.uint8), kernel_dilate, iterations=3)
    
51. 
    
52.     region_dilation_bin = (region_dilation > 0).astype(np.uint8) * 255
    
53.    
    
54.     contours, _ = cv2.findContours(region_dilation_bin, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
    
55.    
    
56.     cv2.drawContours(image_8u, contours, -1, 255, 1)
    
57.    
    
58.     result_file_path = file_path.replace('.png', '_result.png')
    
59.    
    
60.     cv2.imwrite(result_file_path, image_8u)
    
61. 
    
62. print("处理完成")

复制代码

OPENCV HALCON双路高手！

holybang 发表于 2025-10-7 21:01
对应Halcon代码如下：

OpenCV代码如下：

感谢大佬的代码和思路！牛哇！

holybang 发表于 2025-10-7 21:01
对应Halcon代码如下：

OpenCV代码如下：

再请教一下大佬，这种异物与产品灰度差异不是那么大的该怎么处理，无法用之前图像一样的条件筛出来

虎虎虎 发表于 2025-10-13 14:34
再请教一下大佬，这种异物与产品灰度差异不是那么大的该怎么处理，无法用之前图像一样的条件筛出来
...

这个异物位于物体的边缘部分，过滤参数和中心部分的异物不同。所以我先确定了一个边缘的分区之后，再用一组新参数过滤这种异物。Halcon代码已按照这个逻辑更新，结果图也增加了一个。OpenCV的代码暂时没有改。

holybang 发表于 2025-10-15 21:03
这个异物位于物体的边缘部分，过滤参数和中心部分的异物不同。所以我先确定了一个边缘的分区之后，再用一 ...

感谢大佬的回复

虎虎虎 发表于 2025-10-17 09:18
感谢大佬的回复

不客气，你看一下halcon代码的逻辑，照着实现一下opencv的代码应该也不难