Towards optimal scalability in predictive video coding

We propose a method for efficient SNR scalability in predictive video coding which overcomes the known fundamental difficulties due to the prediction loop. The method is generally applicable to any scalable predictive coder but a special emphasis is given in this paper to standard DCT-based video compression. The compression efficiency of the enhancement layer is substantially improved by casting the design of its prediction module within an estimation-theoretic framework. The resulting prediction of a transform coefficient in the enhancement layer is optimal given both the corresponding reconstructed coefficient at the base layer, and the enhancement layer reconstructed coefficient in the previous frame. Simulation shows consistent, substantial improvement in PSNR of the reconstructed enhancement layer frames. For two-layer systems, such optimal prediction can be closely approximated by a simple switched prediction scheme which is of negligible complexity. The proposed method may be easily combined with known temporal scalability methods to provide further improvement in compression performance over a wide range of bit rates