kbarna's profile - activity

I second this request...please fix the email account recovery.

I answer my own question: I was looking for the good thing, only on the bad place. I was mislead of knnMatch (k nearest neighbour) function from the DescriptorMatcher module.

For this problem the classifiers in the ml (machine learning) module should be used instead those from the DescriptorMatcher module.

The ml module contains all the necessary classifiers, from k-nearest neighbors to SVM or the naive bayesian classifier.

Qt works well with OpenGL. Did you try QGLWidget or QOpenGLWidget?

IIRC the QOpenGLWidget is a low-level class that gives you direct access to the OpenGL routines, while QGLWidget is higher level that can use Qt functions for drawing (still with hardware acceleration).

Here is an example: http://doc.qt.io/qt-5/qtopengl-2dpain...

To display an OpenCV Mat in Qt, you have to convert it to QImage (BGR->RGB), then to QPixmap. There are several examples how to do this. If you are using OpenGL routines, I think you can display directly a gpuMat image (but I never tried this).

Hi,

I'm wondering if I can use (and how?) the DescriptorMatcher class to classify other data than keypoints. More precisely I would like to use them for texture classification.

I have N texture classes (C1, C2,...CN), each with K features (let's say the Haralick features: F1...FK). For each texture I have several patches used for training (so texture Ci is described by a [Mi*K] matrix where Miis the number of patches). Now, I would like to classify a patch P in one of those texture classes.

I know I could write my own classifier (KNN, naive Bayesian), but I'm wondering if one of the DescriptorMatcher classes could do the same? The documentation of this class is very sparse and I only found examples classifying keypoint data obtained fromalgorithms like SIFT, SURF, BRIEF...). Or are there other methods for this in OpenCV?

Thanks for any hints!