Commutative Mode-Dependent Transform for Video Intra Coding

In order to further improve intra-coding efficiency, the latest video coding reference software JEM adopts a few intra-coding techniques which extremely increase the coding complexity. In view of this, this paper proposes a commutative mode-dependent transform (CMDT) framework for intra coding, aiming to make a trade-off between the coding efficiency and the coding complexity. In this work, we first design a mode-dependent transform set for each specified intra mode, where only DCT-II and DST-VII are used. Meanwhile, due to the complex and varying textures in natural videos, it is difficult to adapt the various characteristics of the prediction residuals using a set of fixed transform cores. Considering this, we develop a commutative mechanism inspired by the observation that the adj acent intra modes have similar characteristics of the prediction residuals. Through the commutative mechanism, the horizontal and vertical transform cores of a specified intra mode are exchanged as the best transform set of the adj acent intra mode, which could improve the transform coding efficiency without any syntax overhead. The CMDT has been implemented on the reference software JEM 7.0 and experimental results show that the CMDT can achieve 2.11%, 2.53% and 2.59% bit rate saving in average on luma and chroma components, respectively, with negligible encoding and decoding time increase.