otsu算法选择使类间方差最大的灰度值为阈值，具有很好的效果。
1、计算直方图并归一化histogram
2、计算图像灰度均值avgValue.
3、计算直方图的零阶w[i]和一级矩u[i]
4、计算并找到最大的类间方差（between-class variance）
variance[i]=(avgValue*w[i]-u[i])*(avgValue*w[i]-u[i])/(w[i]*(1-w[i]))
对应此最大方差的灰度值即为要找的阈值
5、用找到的阈值二值化图像
这个方法也可以用于图像分割。
void ImageBinarization(IplImage *src) { /*对灰度图像二值化，自适应门限threshold*/ int i,j,width,height,step,chanel,threshold; /*size是图像尺寸，svg是灰度直方图均值，va是方差*/ float size,avg,va,maxVa,p,a,s; unsigned char *dataSrc; float histogram[256]; width = src->width; height = src->height; dataSrc = (unsigned char *)src->imageData; step = src->widthStep/sizeof(char); chanel = src->nChannels; /*计算直方图并归一化histogram*/ for(i=0; i<256; i++) histogram[i] = 0; for(i=0; i<height; i++) for(j=0; j<width*chanel; j++) { histogram[dataSrc[i*step+j]-'0'+48]++; } size = width * height; for(i=0; i<256; i++) histogram[i] /=size; /*计算灰度直方图中值和方差*/ avg = 0; for(i=0; i<256; i++) avg += i*histogram[i]; va = 0; for(i=0; i<256; i++) va += fabs(i*i*histogram[i]-avg*avg); /*利用加权最大方差求门限*/ threshold = 20; maxVa = 0; p = a = s = 0; for(i=0; i<256; i++) { p += histogram[i]; a += i*histogram[i]; s = (avg*p-a)*(avg*p-a)/p/(1-p); if(s > maxVa) { threshold = i; maxVa = s; } } /*二值化*/ for(i=0; i<height; i++) for(j=0; j<width*chanel; j++) { if(dataSrc[i*step+j] > threshold) dataSrc[i*step+j] = 255; else dataSrc[i*step+j] = 0; } }