//! assuming row vectors (a row is a sample)
void cvSoftmax(CvMat * src, CvMat * dst){
CV_FUNCNAME("cvSoftmax");
__BEGIN__;
CV_ASSERT(cvCountNAN(src)<1);
cvExp(src,dst);
CV_ASSERT(cvCountNAN(dst)<1);
const int dtype = CV_MAT_TYPE(src->type);
CvMat * sum = cvCreateMat(src->rows,1,dtype);
CvMat * sum_repeat = cvCreateMat(src->rows,src->cols,dtype);
cvReduce(dst,sum,-1,CV_REDUCE_SUM);
CV_ASSERT(cvCountNAN(sum)<1);
cvRepeat(sum,sum_repeat);
cvDiv(dst,sum_repeat,dst);
cvReleaseMat(&sum);
cvReleaseMat(&sum_repeat);
__END__;
}
//------------------------------------------------------------------------------
// Color Similarity Matrix Calculation
//------------------------------------------------------------------------------
CvMat *colorsim(int nbins, double sigma) {
CvMat *xc=cvCreateMat(1,nbins, CV_32FC1);
CvMat *yr=cvCreateMat(nbins,1, CV_32FC1);
CvMat *x=cvCreateMat(nbins,nbins, CV_32FC1);
CvMat *y=cvCreateMat(nbins,nbins, CV_32FC1);
CvMat *m=cvCreateMat(x->rows,x->rows, CV_32FC1);
// Set x,y directions
for (int j=0;j<nbins;j++) {
cvSetReal2D(xc,0,j,(j+1-0.5)/nbins);
cvSetReal2D(yr,j,0,(j+1-0.5)/nbins);
}
// Set u,v, meshgrids
for (int i=0;i<x->rows;i++) {
cvRepeat(xc,x);
cvRepeat(yr,y);
}
CvMat *sub = cvCreateMat(x->rows,y->cols,CV_32FC1);
cvSub(x,y,sub);
cvAbs(sub,sub);
cvMul(sub,sub,sub);
cvConvertScale(sub,sub,-1.0/(2*sigma*sigma));
cvExp(sub,sub);
cvSubRS(sub,cvScalar(1.0),m);
cvReleaseMat(&xc);
cvReleaseMat(&yr);
cvReleaseMat(&x);
cvReleaseMat(&y);
cvReleaseMat(&sub);
return m;
}
void LogDCT()
{
allocateImages();
if(!cvIm32F) {
cvIm32Fin = cvCreateImage(cvGetSize(cvImGray), IPL_DEPTH_32F, 1);
cvIm32F = cvCreateImage(cvGetSize(cvImGray), IPL_DEPTH_32F, 1);
cvImDCT = cvCreateImage(cvGetSize(cvImGray), IPL_DEPTH_32F, 1);
}
cvConvertScale(cvImGray, cvIm32Fin);
// cvConvertScale(cvImGray, cvIm32F);
cvLog(cvIm32Fin, cvIm32F);
cvDCT(cvIm32F, cvImDCT, CV_DXT_FORWARD);
/* OUTPUTS
*
char * LogDCT_outputnames_list[] = {
"LogDCT",
"Log DCT Cropped",
"Log DCT Inv",
"DCT Inv",
"Final"};
*/
if(LogDCT_output.curitem == 0) // "LogDCT"
{
cvConvertScale(cvImDCT, cvImGray, 100.);
}
// Reduce low DCT
float rad = (float)LogDCT_radius / 100.f;
for(int r = 0; r<cvImDCT->height; r++)
{
float fr = (float)r / (float)cvImDCT->height;
float * line = (float *)(cvImDCT->imageData + r*cvImDCT->widthStep);
for(int c = 0; c<cvImDCT->width; c++)
{
float fc = (float)c / (float)cvImDCT->width;
// float dc = fc*fc;
// float dr = fr*fr;
if(fc > rad || fr > rad)
{
line[c] = 0;
}
}
}
if(LogDCT_output.curitem == 1) // "Log DCT Cropped"
{
cvConvertScale(cvImDCT, cvImGray, 100.);
}
cvDCT(cvImDCT, cvIm32F, CV_DXT_INVERSE);
if(LogDCT_output.curitem == 2) // "LogDCT Inv"
{
cvConvertScale(cvIm32F, cvImGray, 1.);
}
cvExp(cvIm32F, cvImDCT);
//cvConvertScale(cvImDCT, cvImGray, 1.);
//cvConvertScale(cvIm32F, cvImGray, 1.);
cvConvertScale(cvImDCT, cvImDCT, LogDCT_coef);
if(LogDCT_output.curitem == 3) // "DCT Inv"
{
cvConvertScale(cvImDCT, cvImGray, 1.);
}
// Substract low pass image from input image
cvSub(cvIm32Fin, cvImDCT, cvIm32F);
if(LogDCT_output.curitem == 4) // "DCT Inv - scal"
{
cvConvertScale(cvIm32F, cvImGray, 1.);
}
//cvConvertScale(cvImDCT, cvImGray, 1.);
cvAddS(cvIm32F, cvScalarAll(LogDCT_add), cvImDCT);
if(LogDCT_output.curitem == 5) // "Out"
{
cvConvertScale(cvImDCT, cvImGray, 1.);
}
finishImages();
}
/* log_weight_div_det[k] = -2*log(weights_k) + log(det(Sigma_k)))
covs[k] = cov_rotate_mats[k] * cov_eigen_values[k] * (cov_rotate_mats[k])'
cov_rotate_mats[k] are orthogonal matrices of eigenvectors and
cov_eigen_values[k] are diagonal matrices (represented by 1D vectors) of eigen values.
The <alpha_ik> is the probability of the vector x_i to belong to the k-th cluster:
<alpha_ik> ~ weights_k * exp{ -0.5[ln(det(Sigma_k)) + (x_i - mu_k)' Sigma_k^(-1) (x_i - mu_k)] }
We calculate these probabilities here by the equivalent formulae:
Denote
S_ik = -0.5(log(det(Sigma_k)) + (x_i - mu_k)' Sigma_k^(-1) (x_i - mu_k)) + log(weights_k),
M_i = max_k S_ik = S_qi, so that the q-th class is the one where maximum reaches. Then
alpha_ik = exp{ S_ik - M_i } / ( 1 + sum_j!=q exp{ S_ji - M_i })
*/
double CvEM::run_em( const CvVectors& train_data )
{
CvMat* centered_sample = 0;
CvMat* covs_item = 0;
CvMat* log_det = 0;
CvMat* log_weights = 0;
CvMat* cov_eigen_values = 0;
CvMat* samples = 0;
CvMat* sum_probs = 0;
log_likelihood = -DBL_MAX;
CV_FUNCNAME( "CvEM::run_em" );
__BEGIN__;
int nsamples = train_data.count, dims = train_data.dims, nclusters = params.nclusters;
double min_variation = FLT_EPSILON;
double min_det_value = MAX( DBL_MIN, pow( min_variation, dims ));
double likelihood_bias = -CV_LOG2PI * (double)nsamples * (double)dims / 2., _log_likelihood = -DBL_MAX;
int start_step = params.start_step;
int i, j, k, n;
int is_general = 0, is_diagonal = 0, is_spherical = 0;
double prev_log_likelihood = -DBL_MAX / 1000., det, d;
CvMat whdr, iwhdr, diag, *w, *iw;
double* w_data;
double* sp_data;
if( nclusters == 1 )
{
double log_weight;
CV_CALL( cvSet( probs, cvScalar(1.)) );
if( params.cov_mat_type == COV_MAT_SPHERICAL )
{
d = cvTrace(*covs).val[0]/dims;
d = MAX( d, FLT_EPSILON );
inv_eigen_values->data.db[0] = 1./d;
log_weight = pow( d, dims*0.5 );
}
else
{
w_data = inv_eigen_values->data.db;
if( params.cov_mat_type == COV_MAT_GENERIC )
cvSVD( *covs, inv_eigen_values, *cov_rotate_mats, 0, CV_SVD_U_T );
else
cvTranspose( cvGetDiag(*covs, &diag), inv_eigen_values );
cvMaxS( inv_eigen_values, FLT_EPSILON, inv_eigen_values );
for( j = 0, det = 1.; j < dims; j++ )
det *= w_data[j];
log_weight = sqrt(det);
cvDiv( 0, inv_eigen_values, inv_eigen_values );
}
log_weight_div_det->data.db[0] = -2*log(weights->data.db[0]/log_weight);
log_likelihood = DBL_MAX/1000.;
EXIT;
}
if( params.cov_mat_type == COV_MAT_GENERIC )
is_general = 1;
else if( params.cov_mat_type == COV_MAT_DIAGONAL )
is_diagonal = 1;
else if( params.cov_mat_type == COV_MAT_SPHERICAL )
is_spherical = 1;
/* In the case of <cov_mat_type> == COV_MAT_DIAGONAL, the k-th row of cov_eigen_values
contains the diagonal elements (variations). In the case of
<cov_mat_type> == COV_MAT_SPHERICAL - the 0-ths elements of the vectors cov_eigen_values[k]
are to be equal to the mean of the variations over all the dimensions. */
CV_CALL( log_det = cvCreateMat( 1, nclusters, CV_64FC1 ));
CV_CALL( log_weights = cvCreateMat( 1, nclusters, CV_64FC1 ));
CV_CALL( covs_item = cvCreateMat( dims, dims, CV_64FC1 ));
CV_CALL( centered_sample = cvCreateMat( 1, dims, CV_64FC1 ));
CV_CALL( cov_eigen_values = cvCreateMat( inv_eigen_values->rows, inv_eigen_values->cols, CV_64FC1 ));
CV_CALL( samples = cvCreateMat( nsamples, dims, CV_64FC1 ));
CV_CALL( sum_probs = cvCreateMat( 1, nclusters, CV_64FC1 ));
sp_data = sum_probs->data.db;
// copy the training data into double-precision matrix
for( i = 0; i < nsamples; i++ )
{
const float* src = train_data.data.fl[i];
double* dst = (double*)(samples->data.ptr + samples->step*i);
for( j = 0; j < dims; j++ )
dst[j] = src[j];
}
if( start_step != START_M_STEP )
{
for( k = 0; k < nclusters; k++ )
{
if( is_general || is_diagonal )
{
w = cvGetRow( cov_eigen_values, &whdr, k );
if( is_general )
cvSVD( covs[k], w, cov_rotate_mats[k], 0, CV_SVD_U_T );
else
cvTranspose( cvGetDiag( covs[k], &diag ), w );
w_data = w->data.db;
for( j = 0, det = 1.; j < dims; j++ )
det *= w_data[j];
if( det < min_det_value )
{
if( start_step == START_AUTO_STEP )
det = min_det_value;
else
EXIT;
}
log_det->data.db[k] = det;
}
else
{
d = cvTrace(covs[k]).val[0]/(double)dims;
if( d < min_variation )
{
if( start_step == START_AUTO_STEP )
d = min_variation;
else
EXIT;
}
cov_eigen_values->data.db[k] = d;
log_det->data.db[k] = d;
}
}
cvLog( log_det, log_det );
if( is_spherical )
cvScale( log_det, log_det, dims );
}
for( n = 0; n < params.term_crit.max_iter; n++ )
{
if( n > 0 || start_step != START_M_STEP )
{
// e-step: compute probs_ik from means_k, covs_k and weights_k.
CV_CALL(cvLog( weights, log_weights ));
// S_ik = -0.5[log(det(Sigma_k)) + (x_i - mu_k)' Sigma_k^(-1) (x_i - mu_k)] + log(weights_k)
for( k = 0; k < nclusters; k++ )
{
CvMat* u = cov_rotate_mats[k];
const double* mean = (double*)(means->data.ptr + means->step*k);
w = cvGetRow( cov_eigen_values, &whdr, k );
iw = cvGetRow( inv_eigen_values, &iwhdr, k );
cvDiv( 0, w, iw );
w_data = (double*)(inv_eigen_values->data.ptr + inv_eigen_values->step*k);
for( i = 0; i < nsamples; i++ )
{
double *csample = centered_sample->data.db, p = log_det->data.db[k];
const double* sample = (double*)(samples->data.ptr + samples->step*i);
double* pp = (double*)(probs->data.ptr + probs->step*i);
for( j = 0; j < dims; j++ )
csample[j] = sample[j] - mean[j];
if( is_general )
cvGEMM( centered_sample, u, 1, 0, 0, centered_sample, CV_GEMM_B_T );
for( j = 0; j < dims; j++ )
p += csample[j]*csample[j]*w_data[is_spherical ? 0 : j];
pp[k] = -0.5*p + log_weights->data.db[k];
// S_ik <- S_ik - max_j S_ij
if( k == nclusters - 1 )
{
double max_val = 0;
for( j = 0; j < nclusters; j++ )
max_val = MAX( max_val, pp[j] );
for( j = 0; j < nclusters; j++ )
pp[j] -= max_val;
}
}
}
CV_CALL(cvExp( probs, probs )); // exp( S_ik )
cvZero( sum_probs );
// alpha_ik = exp( S_ik ) / sum_j exp( S_ij ),
// log_likelihood = sum_i log (sum_j exp(S_ij))
for( i = 0, _log_likelihood = likelihood_bias; i < nsamples; i++ )
{
double* pp = (double*)(probs->data.ptr + probs->step*i), sum = 0;
for( j = 0; j < nclusters; j++ )
sum += pp[j];
sum = 1./MAX( sum, DBL_EPSILON );
for( j = 0; j < nclusters; j++ )
{
double p = pp[j] *= sum;
sp_data[j] += p;
}
_log_likelihood -= log( sum );
}
// check termination criteria
if( fabs( (_log_likelihood - prev_log_likelihood) / prev_log_likelihood ) < params.term_crit.epsilon )
break;
prev_log_likelihood = _log_likelihood;
}
// m-step: update means_k, covs_k and weights_k from probs_ik
cvGEMM( probs, samples, 1, 0, 0, means, CV_GEMM_A_T );
for( k = 0; k < nclusters; k++ )
{
double sum = sp_data[k], inv_sum = 1./sum;
CvMat* cov = covs[k], _mean, _sample;
w = cvGetRow( cov_eigen_values, &whdr, k );
w_data = w->data.db;
cvGetRow( means, &_mean, k );
cvGetRow( samples, &_sample, k );
// update weights_k
weights->data.db[k] = sum;
// update means_k
cvScale( &_mean, &_mean, inv_sum );
// compute covs_k
cvZero( cov );
cvZero( w );
for( i = 0; i < nsamples; i++ )
{
double p = probs->data.db[i*nclusters + k]*inv_sum;
_sample.data.db = (double*)(samples->data.ptr + samples->step*i);
if( is_general )
{
cvMulTransposed( &_sample, covs_item, 1, &_mean );
cvScaleAdd( covs_item, cvRealScalar(p), cov, cov );
}
else
for( j = 0; j < dims; j++ )
{
double val = _sample.data.db[j] - _mean.data.db[j];
w_data[is_spherical ? 0 : j] += p*val*val;
}
}
if( is_spherical )
{
d = w_data[0]/(double)dims;
d = MAX( d, min_variation );
w->data.db[0] = d;
log_det->data.db[k] = d;
}
else
{
if( is_general )
cvSVD( cov, w, cov_rotate_mats[k], 0, CV_SVD_U_T );
cvMaxS( w, min_variation, w );
for( j = 0, det = 1.; j < dims; j++ )
det *= w_data[j];
log_det->data.db[k] = det;
}
}
cvConvertScale( weights, weights, 1./(double)nsamples, 0 );
cvMaxS( weights, DBL_MIN, weights );
cvLog( log_det, log_det );
if( is_spherical )
cvScale( log_det, log_det, dims );
} // end of iteration process
//log_weight_div_det[k] = -2*log(weights_k/det(Sigma_k))^0.5) = -2*log(weights_k) + log(det(Sigma_k)))
if( log_weight_div_det )
{
cvScale( log_weights, log_weight_div_det, -2 );
cvAdd( log_weight_div_det, log_det, log_weight_div_det );
}
/* Now finalize all the covariation matrices:
1) if <cov_mat_type> == COV_MAT_DIAGONAL we used array of <w> as diagonals.
Now w[k] should be copied back to the diagonals of covs[k];
2) if <cov_mat_type> == COV_MAT_SPHERICAL we used the 0-th element of w[k]
as an average variation in each cluster. The value of the 0-th element of w[k]
should be copied to the all of the diagonal elements of covs[k]. */
if( is_spherical )
{
for( k = 0; k < nclusters; k++ )
cvSetIdentity( covs[k], cvScalar(cov_eigen_values->data.db[k]));
}
else if( is_diagonal )
{
for( k = 0; k < nclusters; k++ )
cvTranspose( cvGetRow( cov_eigen_values, &whdr, k ),
cvGetDiag( covs[k], &diag ));
}
cvDiv( 0, cov_eigen_values, inv_eigen_values );
log_likelihood = _log_likelihood;
__END__;
cvReleaseMat( &log_det );
cvReleaseMat( &log_weights );
cvReleaseMat( &covs_item );
cvReleaseMat( ¢ered_sample );
cvReleaseMat( &cov_eigen_values );
cvReleaseMat( &samples );
cvReleaseMat( &sum_probs );
return log_likelihood;
}
float
CvEM::predict( const CvMat* _sample, CvMat* _probs ) const
{
float* sample_data = 0;
void* buffer = 0;
int allocated_buffer = 0;
int cls = 0;
CV_FUNCNAME( "CvEM::predict" );
__BEGIN__;
int i, k, dims;
int nclusters;
int cov_mat_type = params.cov_mat_type;
double opt = FLT_MAX;
size_t size;
CvMat diff, expo;
dims = means->cols;
nclusters = params.nclusters;
CV_CALL( cvPreparePredictData( _sample, dims, 0, params.nclusters, _probs, &sample_data ));
// allocate memory and initializing headers for calculating
size = sizeof(double) * (nclusters + dims);
if( size <= CV_MAX_LOCAL_SIZE )
buffer = cvStackAlloc( size );
else
{
CV_CALL( buffer = cvAlloc( size ));
allocated_buffer = 1;
}
expo = cvMat( 1, nclusters, CV_64FC1, buffer );
diff = cvMat( 1, dims, CV_64FC1, (double*)buffer + nclusters );
// calculate the probabilities
for( k = 0; k < nclusters; k++ )
{
const double* mean_k = (const double*)(means->data.ptr + means->step*k);
const double* w = (const double*)(inv_eigen_values->data.ptr + inv_eigen_values->step*k);
double cur = log_weight_div_det->data.db[k];
CvMat* u = cov_rotate_mats[k];
// cov = u w u' --> cov^(-1) = u w^(-1) u'
if( cov_mat_type == COV_MAT_SPHERICAL )
{
double w0 = w[0];
for( i = 0; i < dims; i++ )
{
double val = sample_data[i] - mean_k[i];
cur += val*val*w0;
}
}
else
{
for( i = 0; i < dims; i++ )
diff.data.db[i] = sample_data[i] - mean_k[i];
if( cov_mat_type == COV_MAT_GENERIC )
cvGEMM( &diff, u, 1, 0, 0, &diff, CV_GEMM_B_T );
for( i = 0; i < dims; i++ )
{
double val = diff.data.db[i];
cur += val*val*w[i];
}
}
expo.data.db[k] = cur;
if( cur < opt )
{
cls = k;
opt = cur;
}
/* probability = (2*pi)^(-dims/2)*exp( -0.5 * cur ) */
}
if( _probs )
{
CV_CALL( cvConvertScale( &expo, &expo, -0.5 ));
CV_CALL( cvExp( &expo, &expo ));
if( _probs->cols == 1 )
CV_CALL( cvReshape( &expo, &expo, 0, nclusters ));
CV_CALL( cvConvertScale( &expo, _probs, 1./cvSum( &expo ).val[0] ));
}
__END__;
if( sample_data != _sample->data.fl )
cvFree( &sample_data );
if( allocated_buffer )
cvFree( &buffer );
return (float)cls;
}
IplImage*
ComputeSaliency(IplImage* image, int thresh, int scale) {
//given a one channel image
unsigned int size = floor(pow(2,scale)); //the size to do teh saliency @
IplImage* bw_im = cvCreateImage(cvSize(size,size),
IPL_DEPTH_8U,1);
cvResize(image, bw_im);
IplImage* realInput = cvCreateImage( cvGetSize(bw_im), IPL_DEPTH_32F, 1);
IplImage* imaginaryInput = cvCreateImage( cvGetSize(bw_im), IPL_DEPTH_32F, 1);
IplImage* complexInput = cvCreateImage( cvGetSize(bw_im), IPL_DEPTH_32F, 2);
cvScale(bw_im, realInput, 1.0/255.0);
cvZero(imaginaryInput);
cvMerge(realInput, imaginaryInput, NULL, NULL, complexInput);
CvMat* dft_A = cvCreateMat( size, size, CV_32FC2 );
// copy A to dft_A and pad dft_A with zeros
CvMat tmp;
cvGetSubRect( dft_A, &tmp, cvRect(0,0, size,size));
cvCopy( complexInput, &tmp );
// cvZero(&tmp);
cvDFT( dft_A, dft_A, CV_DXT_FORWARD, size );
cvSplit( dft_A, realInput, imaginaryInput, NULL, NULL );
// Compute the phase angle
IplImage* image_Mag = cvCreateImage(cvSize(size, size), IPL_DEPTH_32F, 1);
IplImage* image_Phase = cvCreateImage(cvSize(size, size), IPL_DEPTH_32F, 1);
//compute the phase of the spectrum
cvCartToPolar(realInput, imaginaryInput, image_Mag, image_Phase, 0);
IplImage* log_mag = cvCreateImage(cvSize(size, size), IPL_DEPTH_32F, 1);
cvLog(image_Mag, log_mag);
//Box filter the magnitude, then take the difference
IplImage* log_mag_Filt = cvCreateImage(cvSize(size, size),
IPL_DEPTH_32F, 1);
CvMat* filt = cvCreateMat(3,3, CV_32FC1);
cvSet(filt,cvScalarAll(1.0/9.0));
cvFilter2D(log_mag, log_mag_Filt, filt);
cvReleaseMat(&filt);
cvSub(log_mag, log_mag_Filt, log_mag);
cvExp(log_mag, image_Mag);
cvPolarToCart(image_Mag, image_Phase, realInput, imaginaryInput,0);
cvExp(log_mag, image_Mag);
cvMerge(realInput, imaginaryInput, NULL, NULL, dft_A);
cvDFT( dft_A, dft_A, CV_DXT_INV_SCALE, size);
cvAbs(dft_A, dft_A);
cvMul(dft_A,dft_A, dft_A);
cvGetSubRect( dft_A, &tmp, cvRect(0,0, size,size));
cvCopy( &tmp, complexInput);
cvSplit(complexInput, realInput, imaginaryInput, NULL,NULL);
IplImage* result_image = cvCreateImage(cvGetSize(image),IPL_DEPTH_32F, 1);
double minv, maxv;
CvPoint minl, maxl;
cvSmooth(realInput,realInput);
cvSmooth(realInput,realInput);
cvMinMaxLoc(realInput,&minv,&maxv,&minl,&maxl);
printf("Max value %lf, min %lf\n", maxv,minv);
cvScale(realInput, realInput, 1.0/(maxv-minv), 1.0*(-minv)/(maxv-minv));
cvResize(realInput, result_image);
double threshold = thresh/100.0*cvAvg(realInput).val[0];
cvThreshold(result_image, result_image, threshold, 1.0, CV_THRESH_BINARY);
IplImage* final_result = cvCreateImage(cvGetSize(image),IPL_DEPTH_8U, 1);
cvScale(result_image, final_result, 255.0, 0.0);
cvReleaseImage(&result_image);
//cvReleaseImage(&realInput);
cvReleaseImage(&imaginaryInput);
cvReleaseImage(&complexInput);
cvReleaseMat(&dft_A);
cvReleaseImage(&bw_im);
cvReleaseImage(&image_Mag);
cvReleaseImage(&image_Phase);
cvReleaseImage(&log_mag);
cvReleaseImage(&log_mag_Filt);
cvReleaseImage(&bw_im);
return final_result;
//return bw_im;
}
void CvANN_MLP::calc_activ_func_deriv( CvMat* _xf, CvMat* _df,
const double* bias ) const
{
int i, j, n = _xf->rows, cols = _xf->cols;
double* xf = _xf->data.db;
double* df = _df->data.db;
double scale, scale2 = f_param2;
assert( CV_IS_MAT_CONT( _xf->type & _df->type ) );
if( activ_func == IDENTITY )
{
for( i = 0; i < n; i++, xf += cols, df += cols )
for( j = 0; j < cols; j++ )
{
xf[j] += bias[j];
df[j] = 1;
}
return;
}
else if( activ_func == GAUSSIAN )
{
scale = -f_param1*f_param1;
scale2 *= scale;
for( i = 0; i < n; i++, xf += cols, df += cols )
for( j = 0; j < cols; j++ )
{
double t = xf[j] + bias[j];
df[j] = t*2*scale2;
xf[j] = t*t*scale;
}
}
else
{
scale = -f_param1;
for( i = 0; i < n; i++, xf += cols, df += cols )
for( j = 0; j < cols; j++ )
xf[j] = (xf[j] + bias[j])*scale;
}
cvExp( _xf, _xf );
n *= cols;
xf -= n; df -= n;
// ((1+exp(-ax))^-1)'=a*((1+exp(-ax))^-2)*exp(-ax);
// ((1-exp(-ax))/(1+exp(-ax)))'=(a*exp(-ax)*(1+exp(-ax)) + a*exp(-ax)*(1-exp(-ax)))/(1+exp(-ax))^2=
// 2*a*exp(-ax)/(1+exp(-ax))^2
switch( activ_func )
{
case SIGMOID_SYM:
scale *= -2*f_param2;
for( i = 0; i <= n - 4; i += 4 )
{
double x0 = 1.+xf[i], x1 = 1.+xf[i+1], x2 = 1.+xf[i+2], x3 = 1.+xf[i+3];
double a = x0*x1, b = x2*x3, d = 1./(a*b), t0, t1;
a *= d; b *= d;
t0 = b*x1; t1 = b*x0;
df[i] = scale*xf[i]*t0*t0;
df[i+1] = scale*xf[i+1]*t1*t1;
t0 *= scale2*(2 - x0); t1 *= scale2*(2 - x1);
xf[i] = t0; xf[i+1] = t1;
t0 = a*x3; t1 = a*x2;
df[i+2] = scale*xf[i+2]*t0*t0;
df[i+3] = scale*xf[i+3]*t1*t1;
t0 *= scale2*(2 - x2); t1 *= scale2*(2 - x3);
xf[i+2] = t0; xf[i+3] = t1;
}
for( ; i < n; i++ )
{
double t0 = 1./(1. + xf[i]);
double t1 = scale*xf[i]*t0*t0;
t0 *= scale2*(1. - xf[i]);
df[i] = t1;
xf[i] = t0;
}
break;
case GAUSSIAN:
for( i = 0; i < n; i++ )
df[i] *= xf[i];
break;
default:
;
}
}
void CvANN_MLP::calc_activ_func( CvMat* sums, const double* bias ) const
{
int i, j, n = sums->rows, cols = sums->cols;
double* data = sums->data.db;
double scale = 0, scale2 = f_param2;
switch( activ_func )
{
case IDENTITY:
scale = 1.;
break;
case SIGMOID_SYM:
scale = -f_param1;
break;
case GAUSSIAN:
scale = -f_param1*f_param1;
break;
default:
;
}
assert( CV_IS_MAT_CONT(sums->type) );
if( activ_func != GAUSSIAN )
{
for( i = 0; i < n; i++, data += cols )
for( j = 0; j < cols; j++ )
data[j] = (data[j] + bias[j])*scale;
if( activ_func == IDENTITY )
return;
}
else
{
for( i = 0; i < n; i++, data += cols )
for( j = 0; j < cols; j++ )
{
double t = data[j] + bias[j];
data[j] = t*t*scale;
}
}
cvExp( sums, sums );
n *= cols;
data -= n;
switch( activ_func )
{
case SIGMOID_SYM:
for( i = 0; i <= n - 4; i += 4 )
{
double x0 = 1.+data[i], x1 = 1.+data[i+1], x2 = 1.+data[i+2], x3 = 1.+data[i+3];
double a = x0*x1, b = x2*x3, d = scale2/(a*b), t0, t1;
a *= d; b *= d;
t0 = (2 - x0)*b*x1; t1 = (2 - x1)*b*x0;
data[i] = t0; data[i+1] = t1;
t0 = (2 - x2)*a*x3; t1 = (2 - x3)*a*x2;
data[i+2] = t0; data[i+3] = t1;
}
for( ; i < n; i++ )
{
double t = scale2*(1. - data[i])/(1. + data[i]);
data[i] = t;
}
break;
case GAUSSIAN:
for( i = 0; i < n; i++ )
data[i] = scale2*data[i];
break;
default:
;
}
}
void CLightSet::RunLightPrep(IplImage* src,IplImage* dest)
{
int M,N;
M=0;
N=0;
if (src->roi)
{
M = src->roi->width;
N = src->roi->height;
}
else
{
M = src->width;
N = src->height;
}
CvMat *matD; // create mat for meshgrid frequency matrices
matD = cvCreateMat(M,N,CV_32FC1);
CDM(M,N,matD);
CvMat *matH;
matH = cvCreateMat(M,N,CV_32FC1); // mat for lowpass filter
float D0 = 10.0;
float rH,rL,c;
rH = 2.0;
rL = 0.5;
c = 1.0;
lpfilter(matD,matH,D0,rH,rL,c);
IplImage *srcshift; // shift center
srcshift = cvCloneImage(src);
cvShiftDFT(srcshift,srcshift);
IplImage *log, *temp;
log = cvCreateImage(cvGetSize(src),IPL_DEPTH_32F,1);
temp = cvCreateImage(cvGetSize(src),IPL_DEPTH_32F,1);
cvCvtScale(srcshift,temp,1.0,0);
cvLog(temp,log);
cvCvtScale(log,log,-1.0,0);
CvMat *Fourier;
Fourier = cvCreateMat( M, N, CV_32FC2 );
fft2(log,Fourier);
IplImage* image_im;
image_im = cvCreateImage(cvGetSize(src),IPL_DEPTH_32F,1);
cvSplit(Fourier,dest,image_im,0,0);
cvMul(dest,matH,dest);
cvMul(image_im,matH,image_im);
IplImage *dst;
dst = cvCreateImage(cvGetSize(src),IPL_DEPTH_32F,2);
cvMerge(dest,image_im,0,0,dst);
cvDFT(dst,dst,CV_DXT_INV_SCALE);
cvExp(dst,dst);
cvZero(dest);
cvZero(image_im);
cvSplit(dst,dest,image_im,0,0);
//使得图像按照原来的顺序显示
cvShiftDFT(dest,dest);
double max,min; // normalize
cvMinMaxLoc(dest,&min,&max,NULL,NULL);
cvReleaseImage(&image_im);
cvReleaseImage(&srcshift);
cvReleaseImage(&dst);
cvReleaseImage(&log);
cvReleaseImage(&temp);
cvReleaseMat(&matD);
cvReleaseMat(&matH);
}
