二值化函数cvThreshold（）参数CV_THRESH_OTSU的疑惑【转】

/* otsu_2d：二维最大类间方差阈值分割的快速迭代算法   吴一全 */
#include <iostream>
#include <cv.h>
#include <highgui.h>

using namespace std;
double TwoDimentionOtsu(IplImage *image);
int main()  
{ 
        IplImage* srcImage = cvLoadImage( "E:/image_1/14.bmp",0 ); 
        assert(NULL != srcImage);

        cvNamedWindow("src",0);
        cvShowImage("src",srcImage); 

        clock_t start_time=clock();

        //计算最佳阈值 
        double threshold = TwoDimentionOtsu(srcImage);//70,125

        clock_t end_time=clock();
        cout<< "Running time is: "<<static_cast<double>(end_time-start_time)/CLOCKS_PER_SEC*1000<<"ms"<<endl;//输出运行时间

        cout << "threshold=" << threshold << endl;

        IplImage* biImage = cvCreateImage(cvGetSize(srcImage),8,1);  
        //对图像二值化
        //cvThreshold(srcImage,biImage,255,255, CV_THRESH_OTSU | CV_THRESH_BINARY);
        cvThreshold(srcImage,biImage,threshold,255, CV_THRESH_BINARY);

        cvNamedWindow("binary",0);  
        cvShowImage("binary",biImage);  

        cvWaitKey(0);  

        cvReleaseImage(&srcImage);  
        cvReleaseImage(&biImage);  

        cvDestroyAllWindows();  

        return 0;  
} 
double TwoDimentionOtsu(IplImage *image)
{
        double t0 = 0, s0 = 0, t = 0, s = 0;
        int width = image->width;  
        int height = image->height; 
        double dHistogram[256][256]={0.0};                                //建立二维灰度直方图  
        unsigned long sum0 = 0,sum1 = 0;                                //sum0记录所有的像素值的总和,sum1记录3x3窗口的均值的总和
        uchar* data = (uchar*)image->imageData; 
        for(int i=0; i<height; i++)  
        {  
                for(int j=0; j<width; j++)  
                {  
                        unsigned char nData1 = data[i * image->widthStep + j];//nData1记录当前点（i,j）的像素值
                        sum0 += nData1;
                        unsigned char nData2 = 0;  //nData2记录以当前点（i,j）为中心三领域像素值的平均值
                        int nData3 = 0;                                        //nData3记录以当前点（i,j）为中心三领域像素值之和，注意9个值相加可能超过一个字节  
                        for(int m=i-1; m<=i+1; m++)  
                        {  
                                for(int n=j-1; n<=j+1; n++)  
                                {  
                                        if((m>=0)&&(m<height)&&(n>=0)&&(n<width))
                                                nData3 += data[m * image->widthStep + n];
                                }  
                        }  
                        nData2 = (unsigned char)(nData3/9);    //对于越界的索引值进行补零,邻域均值
                        sum1 += nData2;
                        dHistogram[nData1][nData2]++;  
                }  
        }

        long N = height*width;                //总像素数  
        t = sum0/N;                                        //图像灰度级均值
        s = sum1/N;                                        //邻域平均灰度级的均值

        s0 = s;
        t0 = t; 
        for(int j=0; j<256; j++)
                for(int i=0; i<256; i++) 
                {
                        dHistogram[i][j] = dHistogram[i][j]/N;  //得到归一化的概率分布 
                }

                double w0 = 0.0,w1 = 0.0,u0i = 0.0,u1i = 0.0,u0j = 0.0,u1j = 0.0;

                do
                {
                        t0 = t;
                        s0 = s;
                        w0 = w1 = u0i = u1i = u0j = u1j = 0.0;
                        for (int i = 0,j; i < 256; i++)
                        {
                                for (j = 0; j < s0; j++)
                                {
                                        w0 += dHistogram[i][j];
                                        u0j += dHistogram[i][j] * j;
                                }

                                for (; j < 256; j++)
                                {
                                        w1 += dHistogram[i][j];
                                        u1j += dHistogram[i][j] * j;
                                }

                        }
                        for (int j = 0,i = 0; j < 256; j++)
                        {
                                for (i = 0; i < t0; i++)
                                        u0i += dHistogram[i][j] * i;
                                for (; i < 256; i++)
                                        u1i += dHistogram[i][j] * i;
                        }
                        u0i /= w0;
                        u1i /= w1 ;
                        u0j /= w0;
                        u1j /= w1;

                        t = (u0i + u1i)/2;
                        s = (u0j + u1j)/2;
                }while ( t != t0);//是否可以用这个做为判断条件，有待考究，请高手指点

                return t;//只用t做为阈值，个人也感觉不妥，但没有找到更好的方法，请高手指点

}