Skip to content

O
Opencv

Greenplum / Opencv
11 个月 前同步成功

通知 7
Star 0
Fork 0
O
Opencv

体验新版 GitCode，发现更多精彩内容 >>

提交 583a08d5 编写于 作者: M Maria Dimashova
浏览文件
操作

removed old sample on keypoints matching and added new one

上级 045402e4
隐藏空白更改
内联 并排
Showing with 260 addition and 260 deletion
samples/c/CMakeLists.txt
浏览文件 @ 583a08d5
......@@ -48,7 +48,6 @@ if (BUILD_EXAMPLES)
MY_DEFINE_EXAMPLE(convexhull convexhull.c)
MY_DEFINE_EXAMPLE(delaunay delaunay.c)
MY_DEFINE_EXAMPLE(demhist demhist.c)
MY_DEFINE_EXAMPLE(detectors_sample detectors_sample.cpp)
MY_DEFINE_EXAMPLE(dft dft.c)
MY_DEFINE_EXAMPLE(distrans distrans.c)
MY_DEFINE_EXAMPLE(drawing drawing.c)
......
samples/c/detectors_sample.cpp 已删除 100644 → 0
浏览文件 @ 045402e4
#include <cv.h>
#include <cvaux.h>
#include <highgui.h>
#include <iostream>
using namespace cv;
using namespace std;
inline Point2f applyHomography( const Mat_<double>& H, const Point2f& pt )
{
double w = 1./(H(2,0)*pt.x + H(2,1)*pt.y + H(2,2));
return Point2f( (float)((H(0,0)*pt.x + H(0,1)*pt.y + H(0,2))*w), (float)((H(1,0)*pt.x + H(1,1)*pt.y + H(1,2))*w) );
}
void drawCorrespondences( const Mat& img1, const Mat& img2, const Mat& transfMtr,
const vector<KeyPoint>& keypoints1, const vector<KeyPoint>& keypoints2,
const vector<int>& matches, float maxDist, Mat& drawImg )
{
Scalar RED = CV_RGB(255, 0, 0);
Scalar PINK = CV_RGB(255,130,230);
Scalar GREEN = CV_RGB(0, 255, 0);
Scalar BLUE = CV_RGB(0, 0, 255);
/* Output:
red point - point without corresponding point;
grean point - point having correct corresponding point;
pink point - point having incorrect corresponding point, but excised by threshold of distance;
blue point - point having incorrect corresponding point;
*/
Size size(img1.cols + img2.cols, MAX(img1.rows, img2.rows));
drawImg.create(size, CV_MAKETYPE(img1.depth(), 3));
Mat drawImg1 = drawImg(Rect(0, 0, img1.cols, img1.rows));
cvtColor(img1, drawImg1, CV_GRAY2RGB);
Mat drawImg2 = drawImg(Rect(img1.cols, 0, img2.cols, img2.rows));
cvtColor(img2, drawImg2, CV_GRAY2RGB);
for(vector<KeyPoint>::const_iterator it = keypoints1.begin(); it < keypoints1.end(); ++it )
{
circle(drawImg, it->pt, 3, RED);
}
for(vector<KeyPoint>::const_iterator it = keypoints2.begin(); it < keypoints2.end(); ++it )
{
Point p = it->pt;
circle(drawImg, Point2f(p.x+img1.cols, p.y), 3, RED);
}
Mat vec1(3, 1, CV_32FC1), vec2;
float err = 3;
vector<int>::const_iterator mit = matches.begin();
assert( matches.size() == keypoints1.size() );
for( int i1 = 0; mit < matches.end(); ++mit, i1++ )
{
Point2f pt1 = keypoints1[i1].pt, pt2 = keypoints2[*mit].pt;
Point2f diff = applyHomography(transfMtr, pt1) - pt2;
if( norm(diff) < err )
{
circle(drawImg, pt1, 3, GREEN);
circle(drawImg, Point2f(pt2.x+img1.cols, pt2.y), 3, GREEN);
line(drawImg, pt1, Point2f(pt2.x+img1.cols, pt2.y), GREEN);
}
else
{
/*if( *dit > maxDist )
{
circle(drawImg, pt1, 3, PINK);
circle(drawImg, Point2f(pt2.x+img1.cols, pt2.y), 3, PINK);
}
// TODO add key point filter
else*/
{
circle(drawImg, pt1, 3, BLUE);
circle(drawImg, Point2f(pt2.x+img1.cols, pt2.y), 3, BLUE);
line(drawImg, pt1, Point2f(pt2.x+img1.cols, pt2.y), BLUE);
}
}
}
}
FeatureDetector* createDetector( const string& detectorType )
{
FeatureDetector* fd = 0;
if( !detectorType.compare( "FAST" ) )
{
fd = new FastFeatureDetector( 1/*threshold*/, true/*nonmax_suppression*/ );
}
else if( !detectorType.compare( "STAR" ) )
{
fd = new StarFeatureDetector( 16/*max_size*/, 30/*response_threshold*/, 10/*line_threshold_projected*/,
8/*line_threshold_binarized*/, 5/*suppress_nonmax_size*/ );
}
else if( !detectorType.compare( "SIFT" ) )
{
fd = new SiftFeatureDetector(SIFT::DetectorParams::GET_DEFAULT_THRESHOLD(),
SIFT::DetectorParams::GET_DEFAULT_EDGE_THRESHOLD());
}
else if( !detectorType.compare( "SURF" ) )
{
fd = new SurfFeatureDetector( 400./*hessian_threshold*/, 3 /*octaves*/, 4/*octave_layers*/ );
}
else if( !detectorType.compare( "MSER" ) )
{
fd = new MserFeatureDetector( 5/*delta*/, 60/*min_area*/, 14400/*_max_area*/, 0.25f/*max_variation*/,
0.2/*min_diversity*/, 200/*max_evolution*/, 1.01/*area_threshold*/, 0.003/*min_margin*/,
5/*edge_blur_size*/ );
}
else if( !detectorType.compare( "GFTT" ) )
{
fd = new GoodFeaturesToTrackDetector( 1000/*maxCorners*/, 0.01/*qualityLevel*/, 1./*minDistance*/,
3/*int _blockSize*/, true/*useHarrisDetector*/, 0.04/*k*/ );
}
else
fd = 0;
return fd;
}
DescriptorExtractor* createDescExtractor( const string& descriptorType )
{
DescriptorExtractor* de = 0;
if( !descriptorType.compare( "CALONDER" ) )
{
assert(0);
//de = new CalonderDescriptorExtractor<float>("");
}
else if( !descriptorType.compare( "SURF" ) )
{
de = new SurfDescriptorExtractor( 3/*octaves*/, 4/*octave_layers*/, false/*extended*/ );
}
else
de = 0;
return de;
}
DescriptorMatcher* createDescMatcher( const string& matherType = string() )
{
return new BruteForceMatcher<L2<float> >();
}
const string DETECTOR_TYPE_STR = "detector_type";
const string DESCRIPTOR_TYPE_STR = "descriptor_type";
const string winName = "correspondences";
void iter( Ptr<FeatureDetector> detector, Ptr<DescriptorExtractor> descriptor,
const Mat& img1, float maxDist, Mat& transfMtr, RNG* rng = 0 )
{
if( transfMtr.empty() )
transfMtr = Mat::eye(3, 3, CV_32FC1);
if( rng )
{
transfMtr.at<float>(0,0) = rng->uniform( 0.7f, 1.3f);
transfMtr.at<float>(0,1) = rng->uniform(-0.2f, 0.2f);
transfMtr.at<float>(0,2) = rng->uniform(-0.1f, 0.1f)*img1.cols;
transfMtr.at<float>(1,0) = rng->uniform(-0.2f, 0.2f);
transfMtr.at<float>(1,1) = rng->uniform( 0.7f, 1.3f);
transfMtr.at<float>(1,2) = rng->uniform(-0.1f, 0.3f)*img1.rows;
transfMtr.at<float>(2,0) = rng->uniform( -1e-4f, 1e-4f);
transfMtr.at<float>(2,1) = rng->uniform( -1e-4f, 1e-4f);
transfMtr.at<float>(2,2) = rng->uniform( 0.7f, 1.3f);
}
Mat img2; warpPerspective( img1, img2, transfMtr, img1.size() );
cout << endl << "< Extracting keypoints... ";
vector<KeyPoint> keypoints1, keypoints2;
detector->detect( img1, keypoints1 );
detector->detect( img2, keypoints2 );
cout << keypoints1.size() << " from first image and " << keypoints2.size() << " from second image >" << endl;
if( keypoints1.empty() || keypoints2.empty() )
cout << "end" << endl;
cout << "< Computing descriptors... ";
Mat descs1, descs2;
if( keypoints1.size()>0 && keypoints2.size()>0 )
{
descriptor->compute( img1, keypoints1, descs1 );
descriptor->compute( img2, keypoints2, descs2 );
}
cout << ">" << endl;
cout << "< Matching keypoints by descriptors... ";
vector<int> matches;
Ptr<DescriptorMatcher> matcher = createDescMatcher();
matcher->add( descs2 );
matcher->match( descs1, matches );
cout << ">" << endl;
// TODO time
Mat drawImg;
drawCorrespondences( img1, img2, transfMtr, keypoints1, keypoints2,
matches, maxDist, drawImg );
imshow( winName, drawImg);
}
Ptr<FeatureDetector> detector;
Ptr<DescriptorExtractor> descriptor;
Mat img1;
Mat transfMtr;
RNG rng;
const float maxDistScale = 0.01f;
int maxDist;
void onMaxDistChange( int maxDist, void* )
{
float realMaxDist = maxDist*maxDistScale;
cout << "maxDist " << realMaxDist << endl;
iter( detector, descriptor, img1, realMaxDist, transfMtr );
}
int main(int argc, char** argv)
{
if( argc != 4 )
{
cout << "Format:" << endl;
cout << "./" << argv[0] << " [detector_type] [descriptor_type] [image]" << endl;
return 0;
}
cout << "< Creating detector, descriptor and matcher... ";
detector = createDetector(argv[1]);
descriptor = createDescExtractor(argv[2]);
//Ptr<DescriptorMatcher> matcher = createDescMatcher(argv[3]);
cout << ">" << endl;
if( detector.empty() || descriptor.empty()/* || matcher.empty() */ )
{
cout << "Can not create detector or descriptor or matcher of given types" << endl;
return 0;
}
cout << "< Reading the image... ";
img1 = imread( argv[3], CV_LOAD_IMAGE_GRAYSCALE);
cout << ">" << endl;
if( img1.empty() )
{
cout << "Can not read image" << endl;
return 0;
}
namedWindow(winName, 1);
maxDist = 12;
createTrackbar( "maxDist", winName, &maxDist, 100, onMaxDistChange );
onMaxDistChange(maxDist, 0);
for(;;)
{
char c = (char)cvWaitKey(0);
if( c == '\x1b' ) // esc
{
cout << "Exiting ..." << endl;
return 0;
}
else if( c == 'n' )
iter(detector, descriptor, img1, maxDist*maxDistScale, transfMtr, &rng);
}
waitKey(0);
}
samples/cpp/CMakeLists.txt
浏览文件 @ 583a08d5
......@@ -40,6 +40,7 @@ if (BUILD_EXAMPLES)
MY_DEFINE_EXAMPLE(connected_components connected_components.cpp)
MY_DEFINE_EXAMPLE(contours2 contours2.cpp)
MY_DEFINE_EXAMPLE(keypoints_matching keypoints_matching.cpp)
MY_DEFINE_EXAMPLE(morphology2 morphology2.cpp)
MY_DEFINE_EXAMPLE(segment_objects segment_objects.cpp)
endif(BUILD_EXAMPLES)
......
samples/cpp/keypoints_matching.cpp 0 → 100644
浏览文件 @ 583a08d5
#include <cv.h>
#include <cvaux.h>
#include <highgui.h>
#include <iostream>
using namespace cv;
using namespace std;
inline Point2f applyHomography( const Mat_<double>& H, const Point2f& pt )
{
double z = H(2,0)*pt.x + H(2,1)*pt.y + H(2,2);
if( z )
{
double w = 1./z;
return Point2f( (H(0,0)*pt.x + H(0,1)*pt.y + H(0,2))*w, (H(1,0)*pt.x + H(1,1)*pt.y + H(1,2))*w );
}
return Point2f( numeric_limits<double>::max(), numeric_limits<double>::max() );
}
Mat warpPerspectiveRand( const Mat& src, Mat& dst, RNG* rng )
{
Mat H(3, 3, CV_32FC1);
H.at<float>(0,0) = rng->uniform( 0.8f, 1.2f);
H.at<float>(0,1) = rng->uniform(-0.1f, 0.1f);
H.at<float>(0,2) = rng->uniform(-0.1f, 0.1f)*src.cols;
H.at<float>(1,0) = rng->uniform(-0.1f, 0.1f);
H.at<float>(1,1) = rng->uniform( 0.8f, 1.2f);
H.at<float>(1,2) = rng->uniform(-0.1f, 0.3f)*src.rows;
H.at<float>(2,0) = rng->uniform( -1e-4f, 1e-4f);
H.at<float>(2,1) = rng->uniform( -1e-4f, 1e-4f);
H.at<float>(2,2) = rng->uniform( 0.8f, 1.1f);
warpPerspective( src, dst, H, src.size() );
return H;
}
FeatureDetector* createDetector( const string& detectorType )
{
FeatureDetector* fd = 0;
if( !detectorType.compare( "FAST" ) )
{
fd = new FastFeatureDetector( 10/*threshold*/, true/*nonmax_suppression*/ );
}
else if( !detectorType.compare( "STAR" ) )
{
fd = new StarFeatureDetector( 16/*max_size*/, 5/*response_threshold*/, 10/*line_threshold_projected*/,
8/*line_threshold_binarized*/, 5/*suppress_nonmax_size*/ );
}
else if( !detectorType.compare( "SIFT" ) )
{
fd = new SiftFeatureDetector(SIFT::DetectorParams::GET_DEFAULT_THRESHOLD(),
SIFT::DetectorParams::GET_DEFAULT_EDGE_THRESHOLD());
}
else if( !detectorType.compare( "SURF" ) )
{
fd = new SurfFeatureDetector( 100./*hessian_threshold*/, 3 /*octaves*/, 4/*octave_layers*/ );
}
else if( !detectorType.compare( "MSER" ) )
{
fd = new MserFeatureDetector( 5/*delta*/, 60/*min_area*/, 14400/*_max_area*/, 0.25f/*max_variation*/,
0.2/*min_diversity*/, 200/*max_evolution*/, 1.01/*area_threshold*/, 0.003/*min_margin*/,
5/*edge_blur_size*/ );
}
else if( !detectorType.compare( "GFTT" ) )
{
fd = new GoodFeaturesToTrackDetector( 1000/*maxCorners*/, 0.01/*qualityLevel*/, 1./*minDistance*/,
3/*int _blockSize*/, true/*useHarrisDetector*/, 0.04/*k*/ );
}
else
assert(0);
return fd;
}
GenericDescriptorMatch* createDescriptorMatch( const string& descriptorType )
{
GenericDescriptorMatch* de = 0;
if( !descriptorType.compare( "SIFT" ) )
{
SiftDescriptorExtractor extractor/*( double magnification=SIFT::DescriptorParams::GET_DEFAULT_MAGNIFICATION(),
bool isNormalize=true, bool recalculateAngles=true,
int nOctaves=SIFT::CommonParams::DEFAULT_NOCTAVES,
int nOctaveLayers=SIFT::CommonParams::DEFAULT_NOCTAVE_LAYERS,
int firstOctave=SIFT::CommonParams::DEFAULT_FIRST_OCTAVE,
int angleMode=SIFT::CommonParams::FIRST_ANGLE )*/;
BruteForceMatcher<L2<float> > matcher;
de = new VectorDescriptorMatch<SiftDescriptorExtractor, BruteForceMatcher<L2<float> > >(extractor, matcher);
}
else if( !descriptorType.compare( "SURF" ) )
{
SurfDescriptorExtractor extractor/*( int nOctaves=4,
int nOctaveLayers=2, bool extended=false )*/;
BruteForceMatcher<L2<float> > matcher;
de = new VectorDescriptorMatch<SurfDescriptorExtractor, BruteForceMatcher<L2<float> > >(extractor, matcher);
}
else
assert(0);
return de;
}
void drawCorrespondences( const Mat& img1, const Mat& img2,
const vector<KeyPoint>& keypoints1, const vector<KeyPoint>& keypoints2,
const vector<int>& matches, Mat& drawImg, const Mat& H12 = Mat() )
{
Scalar RED = CV_RGB(255, 0, 0); // red keypoint - point without corresponding point
Scalar GREEN = CV_RGB(0, 255, 0); // green keypoint - point having correct corresponding point
Scalar BLUE = CV_RGB(0, 0, 255); // blue keypoint - point having incorrect corresponding point
Size size(img1.cols + img2.cols, MAX(img1.rows, img2.rows));
drawImg.create(size, CV_MAKETYPE(img1.depth(), 3));
Mat drawImg1 = drawImg(Rect(0, 0, img1.cols, img1.rows));
cvtColor(img1, drawImg1, CV_GRAY2RGB);
Mat drawImg2 = drawImg(Rect(img1.cols, 0, img2.cols, img2.rows));
cvtColor(img2, drawImg2, CV_GRAY2RGB);
// draw keypoints
for(vector<KeyPoint>::const_iterator it = keypoints1.begin(); it < keypoints1.end(); ++it )
{
circle(drawImg, it->pt, 3, RED);
}
for(vector<KeyPoint>::const_iterator it = keypoints2.begin(); it < keypoints2.end(); ++it )
{
Point p = it->pt;
circle(drawImg, Point2f(p.x+img1.cols, p.y), 3, RED);
}
// draw matches
vector<int>::const_iterator mit = matches.begin();
assert( matches.size() == keypoints1.size() );
for( int i1 = 0; mit != matches.end(); ++mit, i1++ )
{
Point2f pt1 = keypoints1[i1].pt,
pt2 = keypoints2[*mit].pt,
dpt2 = Point2f( std::min(pt2.x+img1.cols, float(drawImg.cols-1)), pt2.y);
if( !H12.empty() )
{
if( norm(pt2 - applyHomography(H12, pt1)) > 3 )
{
circle(drawImg, pt1, 3, BLUE);
circle(drawImg, dpt2, 3, BLUE);
continue;
}
}
circle(drawImg, pt1, 3, GREEN);
circle(drawImg, dpt2, 3, GREEN);
line(drawImg, pt1, dpt2, GREEN);
}
}
const string winName = "correspondences";
void doIteration( const Mat& img1, Mat& img2, bool isWarpPerspective, vector<KeyPoint>& keypoints1,
Ptr<FeatureDetector>& detector, Ptr<GenericDescriptorMatch>& descriptor,
double ransacReprojThreshold = -1, RNG* rng = 0 )
{
assert( !img1.empty() );
Mat H12;
if( isWarpPerspective )
{
assert( rng );
H12 = warpPerspectiveRand(img1, img2, rng);
}
else
assert( !img2.empty() && img2.cols==img1.cols && img2.rows== img1.rows );
cout << endl << "< Extracting keypoints from second image..." << endl;
vector<KeyPoint> keypoints2;
detector->detect( img2, keypoints2 );
cout << keypoints2.size() << " >" << endl;
cout << "< Computing and matching descriptors..." << endl;
vector<int> matches;
//if( keypoints1.size()>0 && keypoints2.size()>0 )
{
descriptor->clear();
descriptor->add( img2, keypoints2 );
descriptor->match( img1, keypoints1, matches );
}
cout << ">" << endl;
if( !isWarpPerspective && ransacReprojThreshold >= 0 )
{
cout << "< Computing homography (RANSAC)..." << endl;
vector<Point2f> points1(matches.size()), points2(matches.size());
for( int i = 0; i < matches.size(); i++ )
{
points1[i] = keypoints1[i].pt;
points2[i] = keypoints2[matches[i]].pt;
}
H12 = findHomography( Mat(points1), Mat(points2), CV_RANSAC, ransacReprojThreshold );
cout << ">" << endl;
}
Mat drawImg;
drawCorrespondences( img1, img2, keypoints1, keypoints2, matches, drawImg, H12 );
imshow( winName, drawImg );
}
int main(int argc, char** argv)
{
if( argc != 4 && argc != 6 )
{
cout << "Format:" << endl;
cout << "case1: second image is obtained from the first (given) image using random generated homography matrix" << endl;
cout << argv[0] << " [detectorType] [descriptorType] [image1]" << endl;
cout << "case2: both images are given. If ransacReprojThreshold>=0 then homography matrix are calculated" << endl;
cout << argv[0] << " [detectorType] [descriptorType] [image1] [image2] [ransacReprojThreshold]" << endl;
cout << endl << "Mathes are filtered using homography matrix in case1 and case2 (if ransacReprojThreshold>=0)" << endl;
return 0;
}
bool isWarpPerspective = argc == 4;
double ransacReprojThreshold = -1;
if( !isWarpPerspective )
ransacReprojThreshold = atof(argv[5]);
cout << "< Creating detector, descriptor..." << endl;
Ptr<FeatureDetector> detector = createDetector(argv[1]);
Ptr<GenericDescriptorMatch> descriptor = createDescriptorMatch(argv[2]);
cout << ">" << endl;
if( detector.empty() || descriptor.empty() )
{
cout << "Can not create detector or descriptor or matcher of given types" << endl;
return 0;
}
cout << "< Reading the images..." << endl;
Mat img1 = imread( argv[3], CV_LOAD_IMAGE_GRAYSCALE), img2;
if( !isWarpPerspective )
img2 = imread( argv[4], CV_LOAD_IMAGE_GRAYSCALE);
cout << ">" << endl;
if( img1.empty() || (!isWarpPerspective && img2.empty()) )
{
cout << "Can not read images" << endl;
return 0;
}
cout << endl << "< Extracting keypoints from first image..." << endl;
vector<KeyPoint> keypoints1;
detector->detect( img1, keypoints1 );
cout << keypoints1.size() << " >" << endl;
namedWindow(winName, 1);
RNG rng;
doIteration( img1, img2, isWarpPerspective, keypoints1, detector, descriptor, ransacReprojThreshold, &rng );
for(;;)
{
char c = (char)cvWaitKey(0);
if( c == '\x1b' ) // esc
{
cout << "Exiting ..." << endl;
return 0;
}
else if( isWarpPerspective )
{
doIteration( img1, img2, isWarpPerspective, keypoints1, detector, descriptor, ransacReprojThreshold, &rng );
}
}
waitKey(0);
}
Markdown is supported
0% 或 .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑！
想要评论请 注册 或