A Multi-Objective Optimization Perspective for Joint Consideration of Video Coding Quality

Abstract

Traditional efforts for video coding usually focus on either rate-distortion (R-D) performance or quality smoothness (QS), which may not effectively achieve the desirable visual quality of experience for reconstructed videos. Single-objective optimization cannot guarantee the performance of the other one. In this paper, we would like to introduce a new perspective for the better joint consideration of video coding quality, where both R-D and QS are simultaneously evaluated, and then the traditional video coding optimization problem is converted from single objective to multiple objectives. We provide the details to demonstrate the establishment method of the joint video coding quality (JVCQ), which is given as the comprehensive analytical model to jointly considers spatial and temporal artifacts. The FixedQP results and consumed bits are referred to construct the comparable JVCQ results. Finally, by collecting JVCQ results for four different target bit rates and five different weighting strategies, we can obtain the rate-averaged JVCQ (RJVCQ), weighting-averaged JVCQ (WJVCQ) and rate-weighting-averaged JVCQ (RWJVCQ), respectively. By comparing the different rate control (RC) algorithms, experiments validate the consistency and applicability of the proposed JVCQ with the original two separate evaluation metrics, and the proposed RC algorithms can have better performances. Besides the video coding quality evaluation, the group of JVCQs can be used to guide the optimization process of video coding to achieve gains on both evaluation metrics simultaneously.