TR2012-071

Joint Perceptually-based Intra Prediction and Quantization for HEVC



This paper proposes a new coding scheme which jointly applies perceptual quality metrics to prediction, quantization and rate-distortion optimization (RDO) within the High Efficiency Video Coding (HEVC) framework. A new prediction approach which uses template matching is introduced. The template matching uses a structural similarity metric (SSIM) and a Just-Noticeable Distortion (JND) model. The matched candidates are linearly filtered to generate a prediction. We also modify the JND model and use Supra-threshold Distortion (StD) as the distortion measurement in RDO. Experimental results showing improvements for coding textured areas are presented as well.