ResNet Oriented Fast Mode Decision Algorithm for HEVC Intra Coding

EasyChair Preprint 639

Abstract

The High Efficiency Video Coding (HEVC) standard improves video coding efficiency remarkably at the cost of the highly increased computational complexity. A fast algorithm of coding unit (CU) depth prediction and intra-mode decision was proposed to simplify the CU partitioning and optimal mode decision procedure of intra coding. Firstly, a residual network (ResNet) was trained with the results of HEVC CU depth prediction and intra-prediction mode and the corresponding video sequence. The CU partition process of HEVC is replaced by this trained ResNet that can reduce the coding time effectively. Meanwhile, the process of rough mode selection (RMD) is simplified by the texture direction matching on the current prediction unit (PU) with lower computational complexity. The experimental results show that the proposed algorithm could reduce around 71.2% encoding time with BDBR increased by 1.75% and BDPSNR decreased by 0.093dB on average in all I frame encode scheme compared with HEVC testing model HM-16.7.

Keyphrases: High Efficiency Video Coding (HEVC), ResNet, coding unit (CU) depth prediction, intra prediction, intra-mode decision.

@booklet{EasyChair:639,
  author    = {Da Ai and Yang Gao and Nam Ling and Ying Liu},
  title     = {ResNet Oriented Fast Mode Decision Algorithm for HEVC Intra Coding},
  howpublished = {EasyChair Preprint 639},
  year      = {EasyChair, 2018}}