Statistical analysis of the QTMT structure: intra mode decision

Abstract

As part of their contribution to the continuous development of video technologies, International Telecommunication Union (ITU) and Moving Picture Experts Group (MPEG) created the Joint Expert Team (JVET) in order to study the potential need for the standardization of the future video coding, called Versatile Video Coding (VVC). The VVC project was aiming at outperforming the coding efficiency of the High-Efficiency Video Coding (HEVC) by 40-50% with the same perceptual quality. With the release of its final version in July 2020, VVC has demonstrated a superior coding efficiency, up to 40% [6] of bit reduction beyond HEVC. However, it introduces significant computational complexity in the encoding engine. Such complexity should be analyzed. In this paper, the distribution of intra modes is illustrated through statistical analysis in order to speed up the intra mode decision process in the VVC. Experimental results show that the VVC uses the 6 Most Probable Modes (MPM) as the most used intra modes. In addition, these commonly used intra modes are usually numerous on deeper Multi-type Tree (MT) depths unrelatedly to the Quantization Parameter (QP) but they may vary according to the video texture characteristics and the used intra coding tool, whether Multi Refrence Lines (MRL), Matrix Weighted Intra Prediction (MIP) or Intra Sub Partitions (ISP). This work can be considered as a milestone for proposing intelligent algorithms to perform fast encoding decisions in the intra prediction step.