Machine Learning-Based approaches to reduce HEVC intra coding unit partition decision complexity

Amna Maraoui

doi:10.1007/s11042-021-11678-2

2021-11-05

PRE

DOI：10.1007/s11042-021-11678-2

原文链接

原文求助

摘要

En 中文

The use of machine learning techniques to reduce recent video coding standards complexity such as High Efficiency Video Coding (HEVC) has received prominent attention. In fact, the fascinating HEVC standard coding efficiency gap is performed at the expense of dramatically increasing coding complexity. HEVC adopts a similar block-based hybrid video coding framework its predecessor H.264 Advanced Video Coding (H.264/AVC), but it provides a highly flexible hierarchy of unit representation, which includes three units: coding unit (CU), prediction unit (PU) and transform unit (TU). The recursive splitting of CU is content adaptive, which is one of the biggest differences compared to H.264/AVC. Adopting a large variety of coding unit (CU) sizes, the quadtree partition takes the lion's share of the HEVC encoding complexity, making it ever more challenging to use rigid traditional inference models to predict the efficient coding decisions. In this context, this paper investigates the resulting implications on both coding efficiency and encoding complexity, when using Fuzzy Support Vector Machine (FSVM) and convolutional Neural Network (CNN) models for partitioning in the HEVC intra-prediction. The first approach is an online FSVM-based algorithm designed to predict efficiently the CU partition module. The second one is a deep CNN method founded on a large-scale database of substantial CU partition training. Experimental results reveal that the proposed deep CNN approach, with 66.04% complexity reduction, outperforms the proposed online FSVM approach that achieves 45.33%. However, the FSVM with only 0.067% loss in coding efficiency compared to 1.69% engendered with the CNN, is considered as the approach that performs the best tradeoff between the compression efficiency and the complexity reduction when optimizing the HEVC complexity at all Intra configuration.

Keyword:

HEVC

Deep CNN

Online FSVM

Complexity reduction