Multiple description video coding for P2P systems: A modern design approach.

摘要

Multiple description coding (MDC) is a source coding technique that provides an effective way to mitigate the effects of packet errors/loses by making use of multiple channels. The most attractive application of MDC is perhaps the multiple description video coding (MDVC) in the peer-to-peer (P2P) scenario so as to support simultaneous video streaming to a large population of clients. To this goal, a number of MDVC schemes (both non-scalable and scalable) have been proposed in the past years. However, almost all non-scalable schemes would suffer from the prediction mismatch between the references that are used at the encoder and decoder sides (due to motion compensation); whereas most scalable schemes (involving a base-layer and some enhancement layers) would suffer from the inter-dependency within the enhancement-layer information. In our study, we design the seamless P2P-MDVC system to overcome the two drawbacks mentioned above. We keep a common base-layer in all descriptions and transmit independent enhancement layers in different descriptions. The motion compensation is implemented in the common base-layer so that the mismatch problem has been solved, and the received enhancement information can help improve the video quality. In addition to the aforementioned problems, there are two more important issues in our proposed seamless P2P-MDVC: (1) how to generate well-balanced descriptions, i.e., different descriptions with similar qualities and bit-rates, at the encoder side to satisfy the requirement of P2P network, and (2) how to reconstruct pictures with the "best" quality upon receiving more than one description at the decoder side. To solve the first problem, we develop a new set of unitary transforms that produces an R-D coding performance similar to that of DCT and apply them to code enhancement layers so as to construct well-balanced descriptions. In the meantime, in order to reconstruct better pictures at the decoder side, a technique based on the translation and overlapping in the DCT domain is first proposed and applied to the MDVC system that is built from on a set of staggered quantizers, including the non-scalable MDVC and seamless P2P-MDVC. It is found that the quality gain achieved in this way is rather limited. Then, the total variation (TV) regularized optimization based approach is proposed and used in the same system. It is found that a higher quality gain in both objective (e.g. PSNR-based) and subjective (i.e. visual perception) is obtained. As an extension, this TV-based method is finally applied to the seamless P2P-MDVC system that is based on the newly-developed set of unitary transforms to reconstruct the better video frames, showing a rather remarkable improvement when comparing with the existing methods.