Using curved angular intra-frame prediction to improve video coding efficiency

Abstract

This article presents a new curved-based intra-frame prediction method for current and upcoming video coding standards. Our proposal extends conventional straight-line angular modes found on intra-prediction tools to model curved texture characteristics, enhancing the intra-frame prediction process. Our work targets the High Efficiency Video Coding (HEVC) standard for evaluation, although our curved-based method can be used by any other video coding standard. We model curved intra-frame prediction using an offset-based displacement calculation to each predicted sample. The proposal incurs a small bitstream overhead for transmitting the displacement information, which is offset by encoding efficiency gains. Experimental results demonstrate reduced residual energy; consequently, improving BD-Rate for the tested sequences. Evaluations applying eight curve displacement values show an average BD-Rate reduction of 2.69%, 2.49%, and 0.86% for All-Intra-8, All-Intra 10, and Random-Access configurations, respectively. The proposal allows further BD-Rate improvements, albeit at higher encoding complexity.