Zero keypoints detection using ORB descriptor

Question

I am trying to calculate ORB (Oriented FAST and Rotated BRIEF) features for a database of images. The nexr task is to use a Bag Of Words approach in order to calculate the final features of images. My problem is that in some cases I get 0 keypoints from images of the database (either in ORB or in BRISK implementation). My code is from here.

img = cv2.imread('D:/_DATABASES/clothes_second/striped_141.descr',0)
orb = cv2.ORB()
kp = orb.detect(img,None)
kp, des = orb.compute(img, kp)
img2 = cv2.drawKeypoints(img,kp,color=(0,255,0), flags=0)
plt.imshow(img2),plt.show()

What could be done here, at least orb find one keypoint? How is it possible to use dense sampling for those cases?

Answer 1

You can use a dense feature detector, like the one implemented in C++: http://docs.opencv.org/modules/features2d/doc/common_interfaces_of_feature_detectors.html#densefeaturedetector

The thing is, I'm not sure if that has been ported to python yet. But, since the algorithm is not so hard, you could implement it yourself. Here is the implementation in C++:

void DenseFeatureDetector::detectImpl( const Mat& image, vector<KeyPoint>& keypoints, const Mat& mask ) const
{
    float curScale = static_cast<float>(initFeatureScale);
    int curStep = initXyStep;
    int curBound = initImgBound;
    for( int curLevel = 0; curLevel < featureScaleLevels; curLevel++ )
    {
        for( int x = curBound; x < image.cols - curBound; x += curStep )
        {
            for( int y = curBound; y < image.rows - curBound; y += curStep )
            {
                keypoints.push_back( KeyPoint(static_cast<float>(x), static_cast<float>(y), curScale) );
            }
        }

        curScale = static_cast<float>(curScale * featureScaleMul);
        if( varyXyStepWithScale ) curStep = static_cast<int>( curStep * featureScaleMul + 0.5f );
        if( varyImgBoundWithScale ) curBound = static_cast<int>( curBound * featureScaleMul + 0.5f );
    }

    KeyPointsFilter::runByPixelsMask( keypoints, mask );
}

However, as you will notice, this implementation does not deal with the angle of the keypoints. That can be a problem if your images have rotation.