Saliency-based joint distortion model for 3D video coding

In 3D video system, storing and transmitting the large amount of high definition videos is the main challenge. To further reduce video content related redundant data in 3D video encoded with HEVC, this paper proposes an saliency-based joint distortion model for 3D video encoding, which jointly considers the distortion of texture and depth, as well as synthesis distortion. With this model, two optimization methods are proposed. One method weighs the distortion of each coding unit (CU) according to the saliency information to protect the salient regions. And the other method optimizes the distribution of depth video according to the locality of virtual synthesis distortion (VSD), and weighs the CU-level distortion of texture video and VSD with the saliency information of texture video. Different from existing methods, saliency information based CU-level bit allocation is used in both texture and depth videos. Then we optimize saliency-based joint distortion to minimize the bits while keeping the visual quality the same. The experimental results of visual quality show that the proposed methods have gains on eyetracking PSNR (EWPSNR) with the same bitrate as latest HEVC and an existing saliency encoding method for 3D video. Besides, with the same visual quality, our method costs fewer bits.