Opencv笔记分享---大津(otsu)的使用代码

//Input:
//pTile_pic: 灰度图像，*IplImage
//output:
//im_th: pTile_pic 的otsu阈值
//otsu threshold:
int size = 256;
float range[] = {0, 255};
float *pRange[] = {&range[0]};
CvHistogram *p_imhist = cvCreateHist(1, &size, CV_HIST_ARRAY, pRange);
cvCalcHist(&pTile_pic, p_imhist);
double im_th = icvGetThreshVal_Otsu(p_imhist);

//opencv Source code:
static double
icvGetThreshVal_Otsu( const CvHistogram* hist )
{
double max_val = 0;

CV_FUNCNAME( "icvGetThreshVal_Otsu" );
__BEGIN__;
int i, count;
const float* h;
double sum = 0, mu = 0;
bool uniform = false;
double low = 0, high = 0, delta = 0;
float* nu_thresh = 0;
double mu1 = 0, q1 = 0;
double max_sigma = 0;
if( !CV_IS_HIST(hist) || CV_IS_SPARSE_HIST(hist) || hist->mat.dims != 1 )
CV_ERROR( CV_StsBadArg,
"The histogram in Otsu method must be a valid dense 1D histogram" );
count = hist->mat.dim[0].size;
h = (float*)cvPtr1D( hist->bins, 0 );
if( !CV_HIST_HAS_RANGES(hist) || CV_IS_UNIFORM_HIST(hist) )
{
if( CV_HIST_HAS_RANGES(hist) )
{
low = hist->thresh[0][0];
high = hist->thresh[0][1];
}
else
{
low = 0;
high = count;
}
delta = (high-low)/count;
low += delta*0.5;
uniform = true;
}
else
nu_thresh = hist->thresh2[0];
for( i = 0; i < count; i++ )
{
sum += h;
if( uniform )
mu += (i*delta + low)*h;
else
mu += (nu_thresh[i*2] + nu_thresh[i*2+1])*0.5*h;
}

sum = fabs(sum) > FLT_EPSILON ? 1./sum : 0;
mu *= sum;
mu1 = 0;
q1 = 0;
for( i = 0; i < count; i++ )
{
double p_i, q2, mu2, val_i, sigma;
p_i = h*sum;
mu1 *= q1;
q1 += p_i;
q2 = 1. - q1;
if( MIN(q1,q2) < FLT_EPSILON || MAX(q1,q2) > 1. - FLT_EPSILON )
continue;
if( uniform )
val_i = i*delta + low;
else
val_i = (nu_thresh[i*2] + nu_thresh[i*2+1])*0.5;
mu1 = (mu1 + val_i*p_i)/q1;
mu2 = (mu - q1*mu1)/q2;
sigma = q1*q2*(mu1 - mu2)*(mu1 - mu2);
if( sigma > max_sigma )
{
max_sigma = sigma;
max_val = val_i;
}
}
__END__;
return max_val;
}