MOTION INFORMATION CODING: A RD-OPTIMIZED APPROACH

In this document we consider the problem of operating control of a generic video coder. In this framework we want to introduce a new coding mode for a macroblock, called "split mode" and derived from a region-based motion estimation technique. This new mode is promising because it takes into account the actual shape of moving objects, instead of considering arbitrary square blocks. The proposed approach calls for an integration in a RD-optimized framework and new specifically designed tools for encoding. In this report we describe the framework and the region-based motion estimation technique; we propose novel coding techniques for the segmentation map and the motion information associated, based on a theoretical-information analysis; and we provide extensive experimental results supporting the interest of the proposed method.