Improving dynamic texture recognition by using a color spatio-temporal decomposition

The study of Dynamic Textures (DT) is a recent research topic in the field of video processing. Description and recognition of this phenomena is notoriously a difficult problem but necessary, for example, in video indexation system or video synthesis. The contribution of this paper is to show that it is possible to improve the recognition of a color DT with only a part of its information. In our approach, we propose to split a color image sequence into two components (a geometrical component and a textural component) using the Vectorial Rudin-Osher-Fatemi (VROF) model. The obtained components are used in an application of dynamic texture recognition. The experimental results clearly show that the textural part gives better recognition rates than those obtained with the geometrical part or the original video.