Abstract:

Image binary keypoint descriptors provide an efficient alternative for floatingpoint competitors, as they enable faster processing while requiring less memory. We analyze common binary keypoint descriptors, modify the Fast Retina Keypoint (FREAK) descriptor by using the spatial structure information between keypoints, and propose a multipoint FREAK (MPFREAK) descriptor which improves its feature description ability. As for the slow speed of the nearest neighbor algorithm, we propose a fast feature matching algorithm for hamming space, called MLSH, which modifies the localitysensitive hashing (LSH) algorithm to create a much smaller candidates list. Experimental results show that the proposed feature descriptor is more accurate than other algorithms, and the feature matching algorithm has significant effect and is faster than its competitors.

Key words: binary keypoint descriptor, Fast REtinA Keypoint, localitysensitive hashing, spatial structure, Hamming space