Least-Squares Approximation Surfaces for High Quality Intra-Frame Prediction in Future Video Standards
Ramon Fernandes, Gustavo Sanchez, Rodrigo Cataldo, Luciano Volcan Agostini, Cesar Augusto Missio Marcon · 2018
This paper proposes the Least-Squares Approximation Surfaces (LSAS) for achieving a high intra-frame prediction quality in future video coding standards. LSAS approximates the encoding block by a mathematical expression, whose coefficients are transmitted together with each frame of the video. Software evaluations demonstrate that using a second order polynomial obtained the best tradeoff between PSNR and encoding effort. Its evaluation shows an average PSNR gain from 0.8 dB to 5.0 dB higher than the High-Efficiency Video Coding (HEVC) intra-frame prediction, according to the encoded block size. As a counterpart, the LSAS encoding time varies between -47% and 108% compared to HEVC. LSAS can also be combined with HEVC intra-frame prediction, allowing an average increase in PSNR of 2.9 dB with an increase in the computational effort of 215% compared to the original HEVC intra-frame prediction.