Perceptual video compression based on DCT-Domain foveated JND model

In this paper, a DCT-domain foveated JND model is proposed. The proposal finds the foveation blocks in a frame picture by object movement information and contrast information, and uses block's retina eccentricity and view distance to calculate the foveation modulation factor of every block, which is for scaling spatio-temporal JND. Moreover, the foveated JND is utilized in hybrid video coding framework to reduce bit rate. Experimental results show that the proposed model can save bit rate as much as 9.54%, compared to the same coding scheme with the conventional DCT JND, without degradation of visual quality.