【Halcon】基于watershed_threshold的图像分割

整体等于其部分之和 ——欧几里得
整体大于其部分之和 ——Max Watheimer

---

1.算子

通过阈值实现图像的分水岭算法分割
watersheds_threshold(Image : Basins : Threshold : )

2.原理

第1步：
通过分水岭算法watersheds（）获取图像的盆地。

第2步：
根据第一步分水岭算法分离结果，若盆地部分的灰度**< threshold**，则被合并到一起。设B1和B2分别为相邻盆地的最小灰度值，W为将盆地分割为两个盆地的最小灰度值。则分割结果为：

3.案例

1）边缘振幅+分水岭阈值实现图像分割

dev_set_draw ('fill')
dev_set_line_width (2)
dev_set_colored (12)
read_image (ImageLogo, 'mvtec_logo.png')
dev_close_window ()
get_image_size (ImageLogo, Width, Height)
dev_open_window (0, 0, Width, Height, 'black', WindowHandle)
dev_display (ImageLogo)
disp_continue_message (WindowHandle, 'black', 'true')
stop ()
gauss_filter (ImageLogo, ImageGauss, 9)
sobel_amp (ImageGauss, EdgeAmplitude, 'sum_abs', 3)
disp_continue_message (WindowHandle, 'black', 'true')
stop ()
watersheds_threshold (EdgeAmplitude, Basins, 14)
dev_display (Basins)
disp_continue_message (WindowHandle, 'black', 'true')
stop ()
return ()

其他算子：

• gauss_filter(Image : ImageGauss : Size : )高斯平滑滤波
• sobel_amp(Image : EdgeAmplitude : FilterType, Size : )sobel计算图像边缘幅度

2）距离变换+watersheds_threshold等实现分割

 * In this example, the task is to split the touching pellets.
 * First, a simple thresholding operation yields the region of the pellets
 * in the image. Unfortunately, some of the connected components
 * of this region do not represent single pellets but contain more than one
 * pellet. Therefore, in a second step the watersheds are computed based
 * on the distance transformed pellets regions. The corresponding basins
 * are finally used to split the falsely connected pellets by intersecting the
 * basins with the pellets regions.
 * 
dev_set_draw ('margin')
dev_set_colored (12)
read_image (Image, 'pellets')
dev_close_window ()
get_image_size (Image, Width, Height)
dev_open_window (0, 0, Width, Height, 'black', WindowHandle)
threshold (Image, Region, 105, 255)
 * 计算连通域
connection (Region, ConnectedRegions)
select_shape (ConnectedRegions, SelectedRegions, 'area', 'and', 20, 99999)
dev_display (Image)
dev_display (SelectedRegions)
disp_continue_message (WindowHandle, 'black', 'true')
stop ()
 * 距离变换
distance_transform (SelectedRegions, DistanceImage, 'octagonal', 'true', 380, 350)
convert_image_type (DistanceImage, DistanceImageByte, 'byte')
invert_image (DistanceImageByte, DistanceImageInv)
 * 扩大图像灰度范围【0,255】
scale_image_max (DistanceImageInv, DistanceImageInvScaled)
watersheds_threshold (DistanceImageInv, Basins, 5)
dev_display (DistanceImageInvScaled)
dev_display (Basins)
disp_continue_message (WindowHandle, 'black', 'true')
stop ()
 * 
dev_display (Image)
dev_display (SelectedRegions)
dev_set_color ('blue')
dev_display (Basins)
disp_continue_message (WindowHandle, 'black', 'true')
stop ()
 * 图像求交
intersection (Basins, SelectedRegions, SegmentedPellets)
dev_display (Image)
dev_set_colored (12)
dev_display (SegmentedPellets)
disp_continue_message (WindowHandle, 'black', 'true')
stop ()
return ()

其他算子

• connection(Region : ConnectedRegions : : )计算区域的连通域
• select_shape (ConnectedRegions, SelectedRegions, ‘area’, ‘and’, 20, 99999) 根据区域特征（面积、长度等）提取区域
• distance_transform(Region : DistanceImage : Metric, Foreground, Width, Height : )计算区域的距离变换
• convert_image_type(Image : ImageConverted : NewType : )转换图像类型（byte* / direction* / cyclic* / int1* / int2* / uint2* / int4* / int8 / real* / complex*）
• invert_image(Image : ImageInvert : : )反转图像
  
• scale_image_max(Image : ImageScaleMax : : )将图像灰度范围扩大到【0,255】
• intersection(Region1, Region2 : RegionIntersection : : )计算区域的交集

4.参考

• Halcon官方参考文档

5.附件
末尾分享最近在看的阿里技术分享的《不止代码》，关于程序员的职业进阶、发展、规划等等。
感谢阿里大神们的分享！