Adaptive Vector Length Interpolation for whole-frame error concealment of temporal MDC

The video quality of the conventional single description (streaming) coding (SDC) is vulnerable to transmission errors. Once a video frame is lost, it is likely to cause error propagation to the subsequent received frames, and this increases the difficulty of whole-frame error concealment by its preceding and/or succeeding reference frames. The temporal Multiple Description Coding (T-MDC) is more error resilient than SDC by distributing the risks of transmission at a cost of higher coding bit rate overhead which increases with the number of video descriptions. This cost limits the error resilience of T-MDC in a low bandwidth environment, and a small number of descriptions is preferred, where error concealment is still highly demanded to serve as the final line of defense for video quality. In the literature there have been some algorithms developed for the whole-frame error concealment of SDC and T-MDC. This paper proposes a novel algorithm called AVLI (Adaptive Vector Length Interpolation) for the whole-frame error concealment of 2-description T-MDC. AVLI consists of two component algorithms, and can adaptively select one of them with a smaller sum of absolute difference, leading to a higher video quality of Y-PSNR. One component algorithm adopts non-linear interpolation for the 4 × 4 block based motion vector reconstruction in the lost frame which is bi-directionally weighted by the inverse square of the reference motion vector lengths. The other adopts linear interpolation for the pixel based block reconstruction. Our results show that the proposed outperforms the conventional algorithms by 0.5 to 5 dB for the test cases.