What is the sampling radius used to compute a binary descriptor?

asked 2018-01-25 14:49:38 -0500

Is there a programmatic way of getting (and maybe setting) the radius, in pixels, of the full sampling pattern used to compute a keypoint's descriptor (say, BRISK, AKAZE, etc.)? I see that the keypoints detected using AKAZE are all indented from the edges of the image tile by about 32 pixels (more or less). I am processing large images using a tiling scheme (256x256 pixels per tile). Because of this indentation, there are unmatched regions between adjacent tiles. I would like a way to know ahead of time what sampling radius to expect so that I can request tiles that overlap by at least the sampling radius to insure there are no gaps.

edit retag flag offensive close merge delete