Joint source-channel coding of scalable video

We present a joint source-channel coding scheme developed for transmitting scalable video over a noisy channel. This scheme is based on a 2D+t motion-compensated wavelet decomposition of the video sequence, on a vector quantization and on an optimal linear labelling. In P. Knagenhjelm and E. Agrell's work (IEEE Trans. Information Theory, vol. IT 42, pp. 1139-1151, 1996), it is shown that on binary discrete channels, the channel distortion is minimized by linear labelling. In this paper, we provide a vector quantization method dependent on a linear labelling which minimizes not only the channel distortion but at the same time the distortion of our video source. Moreover, by exploiting a statistical model of the spatio-temporal wavelet coefficients we adapt the vector quantizer to the non-stationarity of the source. Simulation results demonstrate the robustness of this scheme under noisy channel conditions.