阈值分割算子之OSTU算法

Bertha 。 2023-08-17 15:36 147阅读 0赞

1、原理参考：[https://www.cnblogs.com/guopengfei/p/4759569.html][https_www.cnblogs.com_guopengfei_p_4759569.html]

2、公式推导：

![339239-20190827235506048-842404924.png][]

3、同halcon的binary\_threshold (Image, Region, 'max\_separability', 'light', UsedThreshold3)算子。

具体实现如下：

gray_histo (Region, Image, AbsoluteHisto, RelativeHisto)
    get_image_size (Image, Width, Height)
    TotalSize:=Width*Height
    intensity (Region, Image, avgValue, Deviation)//avgValue图像总平均灰度
    MaxDiff:=0.0
    Diff:=0.0 //方差
    Thre:=0
    delta_max:=0
    for i := 0 to 255 by 1
        w0:=0.0
        u0:=0.0
        w1:=0.0
        u1:=0.0
        uo_temp:=0
        u1_temp:=0
        delta_temp:=0
        **背景部分
        for j:=0 to i by 1      
            w0:=w0+AbsoluteHisto[j]
            uo_temp:=uo_temp+j*AbsoluteHisto[j]
        endfor
        **前景部分
        w1:=TotalSize-w0
        u1_temp:=avgValue*TotalSize-uo_temp
        **计算两类的平均灰度
        if(w0==0 or w1==0)
            u0:=0
            u1:=0
        else
            u0:=uo_temp/w0   
            u1:=u1_temp/w1 
        endif 
        **依次找到最大类间方差下的阈值
        delta_temp:=w0*w1*pow((u0-u1),2)
        if(delta_temp>delta_max)
            delta_max := delta_temp
            Thre:=i
        endif  
    endfor
    return ()

优化后的代码实现如下：

gray_histo (Region, Image, AbsoluteHisto, RelativeHisto)
    intensity (Region, Image, avgValue, Deviation)//avgValue图像总平均灰度
    wk:=0.0 //分开后前景像素点数占图像的比例
    uk:=0.0 //分开后前景像素数的平均灰度值
    MaxDiff:=0.0
    Diff:=0.0 //方差
    Thre:=0
    for Index := 0 to 255 by 1
        wk:=wk+RelativeHisto[Index]
        uk:=uk+RelativeHisto[Index]*Index
        if(wk<=0.0 or wk>=1.0)
            Diff:=0
        else
            Diff:=(avgValue*wk-uk)*(avgValue*wk-uk )/(wk*(1-wk))
        endif
        if(Diff>MaxDiff)
            MaxDiff:=Diff
            Thre:=Index
        endif
    endfor
    return ()

转载于:https://www.cnblogs.com/baiyy-daheng/p/11421537.html

[https_www.cnblogs.com_guopengfei_p_4759569.html]: https://www.cnblogs.com/guopengfei/p/4759569.html
[339239-20190827235506048-842404924.png]: /images/20230810/351415d420a14a75807a3839d62de6c2.png