An HEVC-Compliant Perceptual Video Coding Scheme based on Just Noticeable Difference Models

In this paper, we first introduce a spatio-temporal Just Noticeable Difference (JND) model based perceptual video coding (PVC) scheme which is completely compliant with HEVC. We incorporate into HEVC quantization process a JND model that takes into account both the temporal JND characteristics of temporal masking effects and the spatial JND characteristics of the luminance adaptation and contrast masking effects as well as contrast sensitivity. To be HEVC-compliant, the transform coefficients are suppressed based on the spatio-temporal JND model before quantization during HEVC encoding process. The spatio-temporal JND model works in transform domains in the cases of non-transform skip modes, but operates in pixel domains for the transform skip cases in HEVC. To make it effective the JND-based suppression, a distortion compensation factor is also proposed to reflect the perceptual distortion into the ratedistortion optimization (RDO) based encoding process. The proposed spatio-temporal JND-based PVC scheme in HEVC yields remarkable bitrate reductions of maximum 53.37% and average 10.15% with negligible PSNR/SSIM quality loss (average PSNR drop of 1.46dB and average SSIM drop 1.46%), compared to the original HEVC reference software (HM 11.0).