Parameter Based Multi-Objective Optimization of Video CODECs

Introduction

Recently a significant amount of research effort has been focused on optimizing video CODECs, especially within the standardization activities of MPEG/JVT. This is due to the high demand of applications requiring efficient on-demand and real-time video coding, supported by the effective usage of capture, processing and display devices and transmissions mediums, which are practically operational under varying constraints.

To this date many optimization methods have been proposed in literature. They can be broadly classified into two categories, namely,

• •

  Algorithm-based optimizations

• •

  Parameter-based optimizations

The algorithm-based optimization methods focus on the direct performance optimization of a given algorithm. Alternatively, parameter-based optimization methods optimize given objectives through the optimal selection of coding parameters. Whilst the first approach focuses on the optimisation of a CODEC when the algorithms are being developed, the latter enables the optimal operation of a standardized algorithm by the selection of optimal sets of parameters.

With the standardization of H.264, the video optimization research has mainly been focused on this standard. Due to the vast amount of effort that has been put into the algorithmic optimization of H.264 during it’s international standardization, the need at present is a unified framework that is capable of selecting the vast number of coding parameters one can set and specify to obtain the CODEC’s optimum performance under given constraints. Therefore the focus of optimisation subsequent to the standardization activity of any video CODEC is parameter based optimisation.

This chapter presents a multi-objective optimisation strategy for the parameter-based optimisation of H.264 AVC (Advanced Video Coding) standard. Amongst a detailed experimental analysis of the proposed framework the chapter provides a case study that demonstrates the practical use of the approach.