Towards a content-based video retrieval system using wavelet-based signatures

This paper presents two new primitives for representing the content of a video in order to be used in a Content-Based Video Retrieval System. The techniques presented here compute first a multiresolution representation using the Haar transform. Two types of signatures are extracted afterwards, one based on multiresolution global color histograms and the other one based on multiresolution local color histo-rams. The tests performed in the experiments include the recall measure achieved with the proposed primitives using a database composed by 62 videos with 817 shots.