Beyond VVC: Towards Perceptual Quality Optimized Video Compression Using Multi-Scale Hybrid Approaches

Abstract

In this paper, we propose a perceptual quality optimization oriented video compression framework using hybrid approaches. The proposed framework, which is built on the top of recently-published Versatile Video Coding (VVC), contains multi-scale optimized coding techniques. Specifically, three major aspects of efforts from coding unit level to video sequence level have been dedicated to obtain substantial compression efficiency improvement. We first propose a block-level rate-distortion optimization (RDO) method with the consideration of block artifacts removal. Subsequently, we propose frame-level perceptual quality optimized convolutional neural networks for the post-processing of each compressed image, within which the channel attention mechanism has been employed to capture and restore the crucial detail in subjective evaluation. We additionally model the bit allocation as sequence-level dynamic programming problem such that optimal perception and bitrate tradeoff could be obtained. Experimental results show that the proposed method achieves 0.98658 MS-SSIM on the validation set in video track of CLIC-2021.