Dynamic texture synthesis for video compression

TLDR: A dynamic texture based compression scheme is devised for videos for the analysis of motion patterns in a video on the basis of optic flow data and then clusters of different motion patterns are created.

Abstract: In this paper, a dynamic texture based compression scheme is devised for videos. Correspondence analysis is explored for the analysis of motion patterns in a video on the basis of optic flow data and then clusters of different motion patterns are created. Dynamic textures tend to disobey Horn and Schunck's assumption of brightness constancy and hence, optic flow residual is used as an indicator of their presence. The correspondence analysis results and optic flow residual are combined together in a new segmentation scheme. The optic flow data tracks the motion of groups of pixels to generate the flow lines. These flow lines are used in a synthesis scheme for creating an illusion of continuously flowing texture. The integration of this synthesis scheme in the compression format gives us considerable bit stream reduction corresponding to the dynamic texture regions. The scheme is integrated with standard H.264/AVC model, texture regions that do not fall under the above precinct of dynamic textures and non-texture regions are encoded-decoded by the H.264 model directly.