Transformation techniques for future video coding

In this paper we present a number of advanced concepts with respect to the transformation of residual blocks in video coders, which go beyond what is incorporated in today’s video coding standards. Some of these concepts will undoubtedly be adopted as part of future standards. We discuss directional transforms for extrapolation-based prediction schemes, shape-adaptive transformation for object-based coding, and large transform sizes. For the latter we provide in-depth coding efficiency results which clearly illustrate the potential benefit, especially for high definition source material, which will dominate the requirements of tomorrow’s video coding standards. The compression efficiency gains that can be achieved with large block transforms range from 6 to 25%. Finally we also comment on the paradigm of coupling partition and transform areas, a principle which can be applied to block transforms as well as to shape-adaptive transforms for object-based coding.