krammer's profile - activity

Hi

I am trying to compile OpenCV on 64bit ubuntu system having Matlab R2014b(64bit) with gcc 4.9.x.

I use the following cmake parameters:

cmake -D CMAKE_BUILD_TYPE=Release -D CMAKE_INSTALL_PREFIX=/home/siddharth/opencv-master/install -D OPENCV_EXTRA_MODULES=/home/siddharth/opencv-master/opencv_contrib-master -D MATLAB_ROOT_DIR=/home/siddharth/matlab -D WITH_IPP=off ..

But during the configuration, I get the following message:

Matlab: -- mex: /home/siddharth/matlab/bin/mex -- Compiler/generator: Not working (bindings will not be generated)

I have checked that I can compile mex files from within Matlab. All other modules are generated in the build directory except the matlab bindings. I have also seen the question: Bindings to access openCV from MATLAB do not generate. But no working solution is there.

I have put the output of cmake at http://pastebin.com/rWcttkTU

How can I proceed now ? (FYI, I don't have Computer Vision Toolbox in Matlab if that matters)

EDIT: I report same behaviour with both OpenCV-3.0.0 beta and opencv-master(from github).

I compiled opencv 3.0.0 beta. But how can I use KAZE features in opencv ?

did you get any solution ?

I have compiled opencv using the instructions at http://docs.opencv.org/doc/tutorials/...

but there is no support for SIFT or SURF. I searched them in the install directory, but they are not there. I had tried the same instructions with opencv-2.4.9 and nonfree modules did compile. How to achieve the same with opencv 3.0 ?

Is openCV using OpenSIFT. If yes, if I have to modify SIFT code for some of my research, should I use it from within OpenCV or directly use OpenSIFT ?

Hi

I have started studying OpenCV source code. I was just wondering can opencv be setup in an IDE for Development ? It helps in finding related classes and functions, somewhat easier. (I am using QtCreator but any IDE would ease the understanding of code)

PCA-SIFT is very robust, comparable to SIFT and described by lesser dimensions than SIFT. Just wondering why it is not included in OpenCV.

How can I use [PCA-SIFT] with the usual datatypes used in OpenCV for feature matching like Keypoint ?

I am trying to match images using SIFT keypoint detector and BRIEF descriptor. The matching is done using Flann LSHIndexParams. I give one image as input, and it is compared against 20 other images. To store the matched image, I use the drawMatches function. It works perfectly for a few of the images, but after a few iterations gives this error:

OpenCV Error: Assertion failed (i1 >= 0 && i1 < static_cast<int>(keypoints1.size ())) in drawMatches, file C:\OpenCV\opencv\modules\features2d\src\draw.cpp, line 207 terminate called after throwing an instance of 'cv::Exception' what(): C:\OpenCV\opencv\modules\features2d\src\draw.cpp:207: error: (-215) i1 >= 0 && i1 < static_cast<int>(keypoints1.size()) in function drawMatches

My code is as follows:

   //Globals

    #define NO_IMAGES 20
    std::vector<KeyPoint> keypoints;
     cv::Mat descriptors;

    // inside main
    Mat imageSource= imread("1.jpg");
    calculateSourceFeatures(imageC);
    for(int i = 1; i <= NO_IMAGES; i++)
    {            
        Mat image1 = imread(pathImage.arg(i).toStdString());
        Mat image;
        resize(image1, image, Size(256,256)/*new dimensions*/,0, 0, INTER_AREA/* interpolation method*/);
        matchFeatures(imageSource, image,i );                        
    }

  // calculate features of input image
  void calculateSourceFeatures(Mat image)
{
cv::Ptr<FeatureDetector> featureDetector = cv::FeatureDetector::create("SIFT");
featureDetector->detect(image, keypoints);
cv::Ptr<DescriptorExtractor> featureExtractor = cv::DescriptorExtractor::create("BRIEF");
featureExtractor->compute(image, keypoints, descriptors);

}

void matchFeatures( Mat image, Mat target, int imageNo)
{
cv::Mat descriptors1;
std::vector<KeyPoint> keypoints1;

cv::Ptr<FeatureDetector> featureDetector = cv::FeatureDetector::create("SIFT");
cv::Ptr<DescriptorExtractor> featureExtractor = cv::DescriptorExtractor::create("BRIEF");
featureDetector->detect(target, keypoints1);
featureExtractor->compute(target, keypoints1, descriptors1);

cv::Ptr<cv::FlannBasedMatcher> matcher = new cv::FlannBasedMatcher(new     cv::flann::LshIndexParams(5, 24, 2));

std::vector< DMatch > matches;
matcher->match( descriptors, descriptors1, matches );

double max_dist = 0; double min_dist = 100;

for( int i = 0; i < descriptors.rows; i++ )
{ double dist = matches[i].distance;
  if( dist < min_dist ) min_dist = dist;
  if( dist > max_dist ) max_dist = dist;
}
std::vector< DMatch > good_matches;

for( int i = 0; i < descriptors.rows; i++ )
{
  if( matches[i].distance <= 2*min_dist )
  {
      good_matches.push_back( matches[i]);
  }
}


Mat img_matches;
// Error comes here after few iterations
drawMatches( image, keypoints, target, keypoints1,
             good_matches, img_matches, Scalar::all(-1), Scalar::all(-1),
             vector<char>(), DrawMatchesFlags::NOT_DRAW_SINGLE_POINTS );


QString filename = "SIFTBRIEF_" + QString::number(imageNo) + ".jpg";
imwrite(filename.toStdString(), img_matches);

}

I have transformed an image by using affine transform by using the function getAffineTransform() by manually passing 3 points for each image. (i.e. I know the affine transform)

Now, I would like to obtain a homogenous matrix of the transform. How is it possible ? The reason, I am doing it is that I am trying to determine the false positives by extracting the features using SURF and matching using FLANN based Matcher . So, if I can get the homogenous transform between the two images, I can just compare the output of warpperspective on original keypoints and the matched keypoints.

I am using the following code to store Mat(SURF features) to a file

  cv::Mat descriptors;
  featureExtractor->compute(image, keypoints, descriptors);
  FileStorage fs("features.yml", FileStorage::WRITE);
  fs<<"oriimage"<<descriptors;

for reading the file, I use

FileStorage fs;
fs.open("features.yml", FileStorage::READ);
Mat m;
fs["oriimage"] >> m;

The Mat is stored and read as: {3., 6., 0., 0., 0., 0., 3., 2., 0., 0., 34.} //Note the .(dot) at the end of each number

while before storing if I do cout << descriptors, it prints the same numbers without the dots.

What is wrong with my code ?

Hi

I was looking at the descriptor extractor matcher tutorial . There a function is used "evaluateFeatureDetector" which calculates repeatibility and correspondence point. What exactly this function does, I am not able to find any reference in the documentation as well.

thanks. I have tried getAffineTransform(). Is it possible to get homogenous matrix from it ?

Hi

I have been able to transform an image using affine transform and perspective transform using the affine transformation tutorial. I have also been able to rotate an image using getRotationMatrix2D. But how can I translate an image ?