Ask Your Question

Revision history [back]

Feature Detection: WarpPerspective not ok for low res images

I am using WarpPerspective using a matrix obtained from FindHomography to cut out and warp a section of an observed image.

The result is good when the observed image has a high res (1920x1080) but incorrect when using low res (640x360): the cut out image is too big.

I am using KAZE to find keypoints, and BFMatcher.knnMatch to find matches, which I filter and give to FindHomography to get the warp matrix.

What is causing this and how could I fix the low res warp? Any parameters I should look at ?