Texture-based Video Indexing

International audienceWe present a method for indexing and searching video sequences based on textural information. Our method proceeds by first defining and computing texture descriptors relevant to the database at hand. Then a similarity distance is computed between a video sample given by the user and small space-time regions in the sequences. The system returns an ordered set of best matches. We show some results on real sequences, which indicate that this scheme has reasonable complexity and performs well in practice. We believe that these good results stem mainly from a) the introduction of a new texture descriptor based on dynamic local H¨older exponents, and b) the automated adapted choice of the relevant parameters