Exploration of encoding time reduction solutions for intra-frame prediction of VVC encoders

• Main Author: Saldanha, Mário Roberto de Freitas
• Publication Date: 2021
• Format: Doctoral thesis
• Language: por
• Source: Repositório Institucional da UFPel - Guaiaca
• Download full: http://guaiaca.ufpel.edu.br/handle/prefix/8176
• topic: CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO; Computação; Video coding; VVC; Intra-frame prediction; Encoding time saving; Machine learning; Codificação de vídeos; Predição intra-quadro; Redução do tempo de codificação; Aprendizado de máquina

The Versatile Video Coding (VVC) standard was developed to supply the current demand for the upcoming video applications, enabling higher compression performance than previous video coding standards and high versatility to adapt to different types of applications and videos. VVC brings several innovations and enhancements in the coding modules, including larger block sizes, more flexible block partitioning, more angular intra-frame prediction modes, affine motion compensation, and multiple transform selection. Although VVC can provide a high compression performance, these new tools significantly increased the encoding effort. This Thesis proposes timesaving solutions to reduce the encoding effort of the VVC intra-frame prediction. A performance analysis of VVC intra-frame prediction is firstly presented to identify the most time-consuming modules that must be prioritized to accomplish the objective of this work. Subsequently, timesaving solutions based on heuristic and machine learning approaches are presented regarding different intra-frame prediction steps of VVC encoding, including block partitioning of luminance and chrominance samples, mode selection of luminance samples, and the transform selection of luminance samples. It was possible to achieve between 8.5% and 61.3% of encoding time reduction with Bjontegaard Delta Bit Rate (BDBR) ranging from 0.4% to 2.4%. These solutions presented competitive results when compared to the related works.