An efficient error resilient technique for applications of one-way video using transcoding and analysis by synthesis

Abstract

In this paper, we describe an efficient error resilient technique for one-way video transmitted over lossy packet networks. In one-way video applications, video sequences are pre-encoded off-line to be stored in a server without any information on transmission error. Therefore, video encoding algorithm, should be designed without any error resilience or with a hypothetical PLR (packet loss rates) for a certain degree of robustness to be transmitted over error prone environments. However, the PLRs differ according to the various kinds of networks, and even the time varying in a network. A dynamic update of video stream for the adaptive enhancement of error resilience was introduced with a transcoding technique to deal with variable PLRs. An error-sensitivity of each pixel is defined and computed in the transcoder, according to the PLR from network feedback and the effects of error propagations. The error-sensitivity was monitored by the transcoder to modify the video stream for the enhancement of error resilience. The basic error resilience scheme is blocking spatial and temporal error propagation. An analysis by synthesis scheme is introduced for 1) the estimate of accurate distortion at the decoder by exact simulation of error concealment, 2) motion vector re-estimation for blocking temporal error propagation. The optimal mode of each macroblock was decided in the rate-distortion (R-D) framework of the receiver at the given PLR and macroblock location. The performance of the proposed algorithm was evaluated with MPEG-2 video stream over lossy packet network scenario. Experimental results show that the proposed error resilience algorithm outperformed the MPEG-2 TM5 and boundary motion compensated error concealment scheme. It was also shown that the proposed algorithm makes stored one-way video stream applicable for any kinds of error prone networks with variable error rates.