在背景较差的表面进行任何一种检测都是非常不理想的。不过话说回来人能分辨出来的东西代码也能,只不过怎么把原理提取出来。
今日在网上找到一案例:
好图:
坏图:
要求是检测右边的那些斑点。看似非常简单,立马就产生了思路,写的时候猛的发现这个视觉系统的打光不好,有强暗反射的偏差,以及它的背景影响太大。那么前面说过,人能轻松看出来的东西,代码也能。于是想,人判断某个区域的瑕疵是通过瑕疵所在的区域的平均属性来断定的。于是乎,这里就可以用到局部阈值分割。
halcon代码:
dev_set_line_width (3)
for Index := 1 to 2 by 1
read_image (Image, 'C:/Users/斌/Desktop/halcon-项目/瑕疵检测/表面污点检测/'+Index+'.jpg')
rgb1_to_gray (Image, GrayImage)
threshold (GrayImage, Regions, 50, 255)
connection (Regions, ConnectedRegions)
select_shape (ConnectedRegions, SelectedRegions, 'area', 'and', 40000, 99999)
fill_up (SelectedRegions, RegionFillUp)
erosion_rectangle1 (RegionFillUp, RegionErosion, 11, 11)
reduce_domain (GrayImage, RegionErosion, ImageReduced)
crop_domain (ImageReduced, ImagePart)
get_image_size (ImagePart, Width, Height)
dev_close_window ()
dev_open_window (0, 0, Width, Height, 'black', WindowHandle)
dev_display (ImagePart)
*注意,这里截取了一个死区域,是要按需求而定
gen_rectangle1 (ROI_0, 0.936842, 200.429, 177.668, 354.897)
reduce_domain (ImagePart, ROI_0, ImageReduced1)
*mean_image (ImageReduced1, ImageMean1, 3, 3)
median_image (ImageReduced1, ImageMedian, 'circle', 2, 'mirrored')
var_threshold (ImageMedian, Region, 10, 10, 0.1, 22, 'dark')
connection (Region, ConnectedRegions1)
select_shape (ConnectedRegions1, SelectedRegions1, 'area', 'and', 6, 99999)
count_obj (SelectedRegions1, Number)
dev_clear_window ()
dev_display (ImagePart)
set_display_font (3600, 30, 'mono', 'true', 'false')
if (Number > 0)
smallest_circle (SelectedRegions1, Row, Column, Radius)
gen_circle_contour_xld (ContCircle, Row, Column, Radius+2, 0, 6.28318, 'positive', 1)
dev_display (ContCircle)
disp_message (3600, 'NG', 'window', 12, 12, 'red', 'false')
else
disp_message (3600, 'OK', 'window', 12, 12, 'green', 'false')
endif
stop ()
endfor
效果: