Abstract: In order to solve the problem that there are other feature points in the image block without considering one feature point in OSID, and the matching speed of histogram descriptor is slow, an improved binary descriptor based on OSID is proposed. In the process of constructing OSID descriptors, the selection of M is fixed. Therefore, when there are multiple feature points in an image block of a feature point, we propose to try to adapt the value of M, enrich the information contained in the descriptors, and improve the correct matching rate of the algorithm. Meanwhile, the histogram descriptors finally generated by OSID are encoded into binary descriptors, and Hamming distance replaces European distance for image matching, so as to improve the matching speed of the algorithm. The test results on the standard dataset show that the improved OSID has better matching accuracy than the similar descriptors and the original algorithm in complex view changes, image blur and JPEG compression scenarios.

Key words: feature matching, improved OSID, binary descriptor