Delay Cognizant Video Coding

Abstract

We address the coding of video using a substream model of transport. Joint source-channel coding (JSCC) is achieved in the code y segmenting date into QoS substreams , and in the transport by provisioning different QoS attributes for those sbstreams. This paper focuses on JSCD in the delay dimension, which we call delay0-cognizant video coding (DCVC), with the goal of increasing traffic capacity in the transport through traffic smoothing. A DCVC segments data by delay objectives and reconstructs the signal asynchronously at the receiver. Compared to conventional synchronous decoding, subjective quality is enhanced by making the perceptual delay representative of the minimum, rather than the maximum, transport delay. We discuss the motivation, design, and implementation of a delay cognizant video codec. Empirical quantification of the substream delay tolerance using perceptual distortion measures suggests that substream delay variability of ten frames if feasible for the video sequences studied.