Matching Local Invariant Features: How Can Contextual Information Help?

Local invariant features are a powerful tool for finding correspondences between images since they are robust to cluttered background, occlusion and viewpoint changes. However, they suffer the lack of global information and fail to resolve ambiguities that can occur when an image has multiple similar regions. Considering some global information will clearly help to achieve better performances. The question is which information to use and how to use it. While previous approaches use context for description, this paper shows that better results are obtained if contextual information is included in the matching process. We compare two different methods which use context for matching and experiments show that a relaxation based approach gives better results.