Texture and Position Based Multiple Transform for Inter-Predicted Residue Coding

The next generation video coding standard beyond HEVC is in the study. Many efficient coding tools are introduced to the hybrid framework. Among the new coding tools, EMT focuses on transform module by multiple transform modes to improve the transform efficiency. However, EMT is implemented on residual blocks ignoring the details of the residual distribution. Hence, EMT has some drawbacks to be repaired. In this paper, we propose a method, named texture and position based multiple transform (TPBMT), to refine EMT in the framework. What's more, the efficiency of TPBMT is verified by sufficient experimental data. When implemented on JEM7.0, TPBMT attains average 1.26%, 1.66% and 1.62% for Y, U and V, respectively, up to 6.07%, 5.19% and 4.87%.