A perceptual quality metric based rate-quality optimization of H.265/HEVC

An effective video coding should not only remove statistical redundancy but also take into account the characteristic of human visual system. Just noticeable difference (JND) represents the maximum distortion that cannot be perceived in a suitable viewing condition. In this paper, we introduce a foveated JND model combined with visual attention model. Though moving object always attracts viewer´s attention when watching video, viewer could not distinguish every detail of all types of movement. We classify the movement into three types to guide foveated JND model. Then we use the foveated JND model to modify block-based multi-metric fusion (BMMF). In the end, we integrate the modified BMMF metric into traditional hybrid coding scheme. Experimental results show that the proposed model can guarantee better subjective performance at the same bit rate.