I'm using the SIFT algorithm to determine the level of similarity between two images. I managed to get the list of descriptors for image but I cant seem to figure what to do next. My problem is that I have a different number of descriptors and their order is maybe not the same. What is the best way to compare these descriptors?
Thanks.
Basic euclidean distance together with a heuristic that stipulated that a match is only accepted if it is significantly better than the second best match. As advocated by the (co) inventer, Lowe
For each descriptor in image A, find its nearest neighbor in image B. That gives you an approximate correspondance between the descriptors in each image. From there, you need to do some geometry consistency checking to eliminate the many false matches that are generated in the first stage. This can be done using RANSAC.
来源：https://stackoverflow.com/questions/9975537/using-sift-descriptors-to-compare-similarity-between-two-images