Exploring curve-based prediction models for intra-frame prediction

• Ano de defesa: 2021
• Autor(a) principal: Fernandes, Ramon Costi
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Pontifícia Universidade Católica do Rio Grande do Sul; Escola Politécnica; Brasil; PUCRS; Programa de Pós-Graduação em Ciência da Computação
• Palavras-chave em Português: Intra-Frame Prediction; Video Coding; Predictive Coding; Predição Intra-Quadros; Codificação de Vídeos; Codificação Preditiva; CIENCIA DA COMPUTACAO::TEORIA DA COMPUTACAO
• Link de acesso: http://tede2.pucrs.br/tede2/handle/tede/9589

This Thesis presents two novel intra-frame prediction models for video coding: Least-Squares Approximation Surfaces (LSAS) and curve-based prediction models. Both approaches enhance intra-frame prediction capabilities by offering a better approximation of encoded block contents. While the LSAS implementation is currently not practical due to its prohibitive bitrate requirements, the curve-based prediction models show promising results. The evaluations for the novel intra-frame prediction models adopt the High Efficiency Video Coding (HEVC) standard, which culminates years of advancements in video coding technologies. Compared to its predecessor, H.264, HEVC achieves up to 50% coding efficiency improvements. This efficiency gain translates into half the encoded video size while keeping the same visual quality. Among the many improvements of HEVC, its intra-frame predictor comes with more prediction modes, capable of modeling more textures in prediction blocks than its predecessors. Improving intra-frame prediction is an essential aspect of the encoding flow, as a better prediction translates into reduced residual energy, consequently improving coding efficiency. All thirty-three angular modes in HEVC received a curve displacement calculation to each predicted sample so that the resulting prediction block models image regions with curved textures. The proposal includes a small overhead in the bitstream syntax elements to transmit the curve displacement value and increased encoding times. However, the enhanced prediction offsets this overhead with improved coding efficiency. The experimental results demonstrate increased prediction accuracy with lower residual energy when applying the Centerline and Radial curve-based models for the displacement calculations. These models achieve an average reduction of the Bjøntegaard-Delta bitrate (BD-Rate) of 2% and 3% for the HEVC test sequences using the All-Intra-8 configuration, for the Centerline and Radial models, respectively, albeit at an increase in encoding times. Higher encoding efficiency is achievable with the combination of both curve models, reaching an average 4% BD-Rate reduction with the All-Intra-8 configuration.