Improving the intra-prediction of H.264 and H.265 video coding standards using adaptive weighted least squares based predictor

Abstract

Advanced Video Coding (H.264/AVC) and its recent extension High Efficiency Video Coding (H.265/HEVC) are the current industrial video coding standards that have directional and planar intra-predictors. The intra-prediction scheme of both standards is dominated by directional predictors. Ordinary least squares (OLS) based adaptive predictor is superior around edge pixels without considering the direction of an edge explicitly. In OLS based predictor, each pixel in the local area contributes equal weight to estimate prediction parameters. We observed that this does not hold true due to noise and the existence of different correlations between the causal context of each neighboring pixel. To address this problem, we proposed an adaptive weighted least squares (AWLS) based predictor that assigns different weights to neighboring pixels to reflect its relative contribution. Experimental results show that the proposed method outperforms the OLS based intraprediction for directional images and marginal improvements are obtained for other tested images and videos.