opencv
| Table of Contents |
Snippets
反転？
まず avi ファイルを読み込んだ時点で。
img = cvQueryFrame( cap ); #if defined(WIN32) || defined(WIN64) img->origin = 0; cvFlip( img, img, 0 ); #endif
いや、
img = cvQueryFrame( cap ); if( img->origin != IPL_ORIGIN_TL ) cvFlip( img, img ); // prevent flipping on Windows
こうか？いや、やはり上のほうが確実っぽい
cvCvtColor, cvCopy などを呼ぶと WIN32 では画像が反転する模様
cvCvtColor( src, dist, CV_BGR2GRAY );
#if defined(WIN32) || defined(WIN64)
dist->origin = img->dist;
#endif
カナ？
CVAPI
CVAPI(int) などの CVAPI は cxcore/include/cxtypes.h にある
extern "C" __declspec(dllexport) int __cdecl
みたいな形になる模様。環境依存配慮
ゴミ
CvEM を GmmPdf として使いたい
うーん、ダメだ。
#ifdef _MSC_VER
#pragma warning(disable:4996)
#pragma comment(lib, "cv.lib")
#pragma comment(lib, "cxcore.lib")
#pragma comment(lib, "cvaux.lib")
#pragma comment(lib, "highgui.lib")
#pragma comment(lib, "ml.lib")
#endif
#include "cv.h"
#include "cvaux.h"
#include "ml.h"
#ifndef CV_EMEXT_INCLUDED
#define CV_EMEXT_INCLUDED
CvEM::CvEM( const CvMat* samples, const CvMat* sample_idx, CvEMParams params, CvMat* labels )
{
int D = params.means->cols;
CvVectors train_data;
train_data.count = params.weights->rows;
train_data.dims = D;
set_params( params, train_data );
}
/*
// Example)
// const int D = 2;
// const int N = 3;
// const int K = 2;
//
// double vs[] = { 3, 3,
// 4, 4,
// 5, 5 }; // row vectors
// double ms[] = { 3, 3,
// 5, 5 }; // row vectors
// double cs0[] = { 1, 0,
// 0, 1 };
// double cs1[] = { 1, 0.1,
// 0.1, 1 };
// double ws[] = { 0.5, 0.5 };
//
// CvMat vecs = cvMat(N, K, CV_64FC1, vs);
// CvMat means = cvMat(K, D, CV_64FC1, ms);
// CvMat **covs = (CvMat**)cvAlloc( K * sizeof(*covs) );
// covs[0] = &cvMat(D, D, CV_64FC1, cs0);
// covs[1] = &cvMat(D, D, CV_64FC1, cs0);
// CvMat weights = cvMat(1, K, CV_64FC1, ws);
// CvMat *probs = cvCreateMat(N , K, CV_64FC1);
// CvEM em;
// em.GmmPdf( &vecs, &means, covs, &weights, probs );
// cvMatPrint( probs );
// cvReleaseMat( &probs );
// cvFree( &covs );
//
// @param CvMat* samples N x D sample vector
// @param CvMat* means K x D mean vector for each cluster
// @param CvMat** covs K x (D x D) covariance matrix for each cluster
// @param CvMat* weights 1 x K weights
// @param CvMat* probs N x K or N x 1 computed probabilities.
// If N x K, probs for each cluster is stored instead of sum of them ( N x 1 )
*/
void CvEMGmmPdf( const CvMat* samples, const CvMat* means, CvMat** covs, const CvMat* weights, CvMat* probs )
{
int N = samples->rows;
int D = samples->cols;
int K = means->rows;
int type = samples->type;
CvEMParams params;
params.covs = (const CvMat**)covs;
params.means = means;
params.weights = weights;
params.probs = probs;
params.nclusters = K;
params.cov_mat_type = CvEM::COV_MAT_GENERIC;
//CvEM em_model( NULL, NULL, params, NULL );
CvMat* sample = cvCreateMat( 1, D, type );
CvMat* prob = cvCreateMat( 1, probs->cols, type );
for( int n = 0; n < N; n++ )
{
cvGetRow( samples, sample, n );
//em_model.predict( sample, prob );
for( int k = 0; k < probs->cols; k++ )
{
cvmSet( probs, n, k, cvmGet( prob, 0, k ) );
}
}
cvReleaseMat( &sample );
cvReleaseMat( &prob );
}
#endif
Mask And
matlab なら no-loop のほうが短くなるのだが・・・・
IplImage* imgB = cvCreateImage( cvGetSize(img), img->depth, 1 );
IplImage* imgG = cvCreateImage( cvGetSize(img), img->depth, 1 );
IplImage* imgR = cvCreateImage( cvGetSize(img), img->depth, 1 );
IplImage* tmpMask = cvCreateImage( cvGetSize(img), IPL_DEPTH_8U, 1 );
cvSplit( img, imgB, imgG, imgR, NULL );
cvSet( mask, cvScalarAll(1) );
cvThreshold( imgR, tmpMask, 95, 1, CV_THRESH_BINARY );
cvAnd( tmpMask, mask, mask );
cvThreshold( imgG, tmpMask, 40, 1, CV_THRESH_BINARY );
cvAnd( tmpMask, mask, mask );
cvThreshold( imgB, tmpMask, 20, 1, CV_THRESH_BINARY );
cvAnd( tmpMask, mask, mask );
IplImage* maxRGB = cvCreateImage( cvGetSize(img), IPL_DEPTH_8U, 1 );
IplImage* minRGB = cvCreateImage( cvGetSize(img), IPL_DEPTH_8U, 1 );
cvMax( imgR, imgG, maxRGB );
cvMax( imgB, maxRGB, maxRGB );
cvMin( imgR, imgG, minRGB );
cvMin( imgB, minRGB, minRGB );
cvAbsDiff( maxRGB, minRGB, tmpMask );
cvThreshold( tmpMask, tmpMask, 15, 1, CV_THRESH_BINARY );
cvAnd( tmpMask, mask, mask );
cvRelease( &maxRGB );
cvRelease( &minRGB );
cvAbsDiff( imgR, imgG, tmpMask );
cvThreshold( tmpMask, tmpMask, 15, 1, CV_THRESH_BINARY );
cvAnd( tmpMask, mask, mask );
cvCmp( imgR, imgB, tmpMask, CV_CMP_GT );
cvAnd( tmpMask, mask, mask );
cvCmp( imgR, imgG, tmpMask, CV_CMP_GT );
cvAnd( tmpMask, mask, mask );
cvReleaseImage( &imgB );
cvReleaseImage( &imgG );
cvReleaseImage( &imgR );
cvReleaseImage( &tmpMask );
YCbCRMAHAL Skincolor
//case CV_SKINCOLOR_YCrCbMAHAL:
//[2] Son Lam Phung; Bouzerdoum, A.; Chai, D., "A novel skin color model in YCbCr color space and its application to human face detection," Image Processing. 2002. Proceedings. 2002 International Conference on , vol.1, no., pp. I-289-I-292 vol.1, 2002
/*
#define MAX_DATA 500
#define DEMIENSION 3
int count; //データの個数
double distance; //マハラノビス距離
CvMat *data[MAX_DATA]; //クラス
CvMat *result = cvCreateMat( DIMENSION, DIMENSION, CV_32F); //共分散行列
CvMat *inversecovar = cvCreateMat( DIMENSION, DIMENSION, CV_32F); //逆共分散行列
CvMat *average = cvCreateMat( DIMENSION, 1, CV_32F); //平均ベクトル
CvMat *input = cvCreateMat( DIMENSION, 1, CV_32F); //比較するベクトル
//クラスの読み込み
//この場合dataの各要素は高さDIMENSION，幅1のベクトル
count = ReadData( data, MAX_DATA);
ReadInput( input); //↓クラスから共分散行列を計算
cvCalcCovarMatrix( data, count, result, average, CV_COVAR_NORMAL | CV_COVAR_SCALE);
cvInvert( result, inversecovar, CV_SVD); //逆行列計算
distance = cvMahalanobis( average, input, inversecovar); //計算した逆共分散行列からマハラノビス距離を計算
*/
// // Rcb \in [75,135], Rcr \in [130,180]
// // (75,118,143), (138,110,150), (184,111,147)
// double m1[] = { 75, 118, 143 };
// double m2[] = { 138, 110, 150 };
// double m3[] = { 184, 111, 147 };
// CvMat mean1 = cvMat( 3, 1, CV_64FC1, m1 );
// CvMat mean2 = cvMat( 3, 1, CV_64FC1, m2 );
// CvMat mean3 = cvMat( 3, 1, CV_64FC1, m3 );
// CvMat cluster_data[img->nSize];
// CvMat *cov1 = cvCreateMat( 3, 3, CV_64FC1 );
// CvMat *cov2 = cvCreateMat( 3, 3, CV_64FC1 );
// CvMat *cov3 = cvCreateMat( 3, 3, CV_64FC1 );
// IplImag* imgYCrCb = cvCreateImage( cvGetSize(img), img->depth, 1 );
// cvCvtColor( img, imgYCrCb, CV_BGR2YCrCb );
// float thresh1 = 255/3.0;
// float thresh2 = 2 * thresh1;
// IplImag* cluster_mask1 = cvCreateImage( cvGetSize(img), IPL_DEPTH_8U, 1 );
// IplImag* cluster_mask2 = cvCreateImage( cvGetSize(img), IPL_DEPTH_8U, 1 );
// IplImag* cluster_mask3 = cvCreateImage( cvGetSize(img), IPL_DEPTH_8U, 1 );
// cvZero( cluster1 ); cvZero( cluster2 ); cvZero( cluster3 );
// int nCluster1 = 0, nCluster2 = 0, nCluster3 = 0;
// int num = 0;
// double v[3];
// for( int i = 0; i < img->height; i++ )
// {
// for( int j = 0; j < img->width; j++ )
// {
// v[0] = imgYCrCb->imageData[img->widthStep * i + j * 3];
// v[1] = imgYCrCb->imageData[img->widthStep * i + j * 3 + 1];
// v[2] = imgYCrCb->imageData[img->widthStep * i + j * 3 + 2];
// if( v[0] < thresh1 )
// {
// cluster1->imageData[img->widthStep * i + j] = 1;
// cluster_data[n++] = cvMat( 3, 1, CV_64FC1, v );
// nCluster1++;
// }
// else if( v[0] < thresh 2)
// {
// cluster2->imageData[img->widthStep * i + j] = 1;
// nCluster2++;
// }
// else
// {
// cluster3->imageData[img->widthStep * i + j] = 1;
// nCluster3++;
// }
// }
// }
// cvCalcCovarMatrix( &cluster_data, nCluster1, cov1, mean1, CV_COVAR_NORMAL | CV_COVAR_SCALE);
// CvMat *invcov1 = cvCreateMat( 3, 3, CV_64FC1 );
// CvMat *invcov2 = cvCreateMat( 3, 3, CV_64FC1 );
// CvMat *invcov3 = cvCreateMat( 3, 3, CV_64FC1 );
// cvInvert( cov1, invcov1, CV_SVD);
// cvInvert( cov2, invcov2, CV_SVD);
// cvInvert( cov3, invcov3, CV_SVD);
// cvReleaseMat( &cov1 );
// cvReleaseMat( &cov2 );
// cvReleaseMat( &cov3 );
// for( int i = 0; i < mat->rows; i++ )
// {
// for( int j = 0; j < mat->cols; j++ )
// {
// double dist1 = cvMahalanobis( mean1, cvmGet( data, i, j ), invcov1 );
// double dist2 = cvMahalanobis( mean2, cvmGet( data, i, j ), invcov2 );
// double dist3 = cvMahalanobis( mean3, cvmGet( data, i, j ), invcov3 );
// if( min( dist1, min( dist2, dist3 ) ) < thresh )
// {
// mask->imageData[img->widthStep * i + j] = 1;
// }
// }
// }
// for( int y = 0; y < img->height; y++ )
// {
// for( int x = 0; x < img->width; x++ )
// {
// v[0] = imgYCrCb->imageData[img->widthStep * y + x * 3];
// v[1] = imgYCrCb->imageData[img->widthStep * y + x * 3 + 1];
// v[2] = imgYCrCb->imageData[img->widthStep * y + x * 3 + 2];
// if( ( v[2] < 130 || v[2] > 180 ) || ( v[1] < 75 || v[1] > 135 ) )
// {
// mask->imageData[img->widthStep * y + x] = 0;
// continue;
// }
//
// }
// }
// double v[3];
// cvSet( mask, cvScalarAll(1) );
// for( int y = 0; y < img->height; y++ )
// {
// for( int x = 0; x < img->width; x++ )
// {
// v[0] = imgYCrCb->imageData[img->widthStep * y + x * 3];
// v[1] = imgYCrCb->imageData[img->widthStep * y + x * 3 + 1];
// v[2] = imgYCrCb->imageData[img->widthStep * y + x * 3 + 2];
// if( ( v[2] < 130 || v[2] > 180 ) || ( v[1] < 75 || v[1] > 135 ) )
// {
// mask->imageData[img->widthStep * y + x] = 0;
// continue;
// }
//
// }
// }
