SVM-based Inter Prediction Mode Decision for HEVC

Abstract

Over the recent years, machine learning technology has witnessed a big progress in many disciplines, particularly in video processing and computer vision. The latest video coding standard is the High Efficiency Video Coding (HEVC). HEVC doubles the video compression ratio at the same video quality and without any degradation, compared to its predecessor, H.264/AVC. However, this high performance is reached by increasing the HEVC encoding complexity, mainly due to its coding structure, named quadtree. That's why HEVC complexity reduction presents a crucial subject. To reduce this complexity, this paper proposes a fast Coding Unit algorithm based on Support Vector Machine (SVM). Simulation results prove that the proposed approach can reduce the computational complexity by 52.4% with 1.62 % BitRate degradation and a negligible loss in terms of PSNR using Low-Delay P configuration. However, the proposed scheme achieves a best tradeoff between both complexity reduction and RD performance, comparing to the earliest HEVC framework.