Perceptual video coding based on saliency and Just Noticeable Distortion for H.265/HEVC

In this paper, a novel Saliency-JND (S-JND) model has been proposed which combines the visual attention model and the visual sensitivity model together. The visual attention model can discover the saliency region where our eyes focus when viewing a video. The visual sensitivity model can define a threshold for unperceived distortions. More perceptual redundancy can be removed by using visual saliency to adjust the JND threshold. For salient regions, the threshold will be decreased based on the saliency values, while the threshold will be increased for non-salient regions. In order to evaluate our proposed S-JND model, we incorporate the model into HEVC software HM 14.0 via the residue adaptive filter process. The experimental results show that our proposed S-JND model obtained 8.16%~28.97% bitrate reduction for HEVC while having nearly the same subjective perceptual quality.