Color pattern diffusion : Verification for wide baseline matching

After image retrieval based on wide baseline match, verification is required since there are still many false match pairs. The popular verification method, geometry, is limited for example in a weak match situation where the number of match pairs is less than enough. In order to improve retrieval results in such cases, this paper puts forward color pattern diffusion verification. Firstly, HSI Color moments in the original region where two corresponding SIFT feature points are detected are extracted from both the query image and the database image and then matched with each other. Secondly, color moments in both diffusion regions are extracted and matched as well. Finally, if the matching similarity of above color moments keeps invariant with color pattern diffusion in image space, the two interest points is taken as a real corresponding pair. Experiments demonstrated this approach works better with near real-time performance in many practical situations, especially in the weak match situation context.