AVS encoding optimization with perceptual just noticeable distortion model

Abstract

Integrating efficient visual perceptual cues into standardized video coding framework can improve performance significantly. In this paper we propose to enhance AVS encoder by using the latest just noticeable distortion (JND) model to adjust DCT coefficients of prediction residues in a content adaptive way. To better modeling JND profile in AVS integer DCT domain, we further derive the JND mapping from the classical DCT domain to AVS Integer DCT domain. The experiment shows that the proposed algorithm can reduce the bitrate by about 13% on average, compared to the AVS standard encoder at similar visual quality.