Filter Banks for Prediction-Compensated Multiple Description Coding

In this paper, a prediction-compensated multiple description (MD) coding framework for two-band filter banks is proposed, in which the coefficients in each subband are split into two descriptions. Each description also includes the prediction residuals of the data in the other description. The designs of the optimal orthogonal and biorthogonal filter banks are formulated in a unified framework, and both one-level and multiple-level decompositions are analyzed. Contrary to the existing MD filter banks in the literature, the optimal filter banks in the proposed scheme are quite similar to those in single description coding. Therefore, the method can be applied to systems with single-description-optimized filter banks and still attain near-optimal performance. Image coding results show that this method achieves better performance and lower complexity than the latest JPEG 2000 based MDC.

I. Introduction

Multiple description coding (MDC) [1] is an attractive technique of combating transmission errors. In MDC, the source signal is encoded into several coded streams called descriptions, which are sent to the receiver via different network paths. Judiciously designed redundancies are introduced in all descriptions such that the reconstruction quality degrades gracefully when some of them are lost. In this paper, we focus on MDC with two descriptions and three decoders, where the decoder receiving one and two descriptions is called the side decoder and the central decoder, respectively. In particular, we are interested in wavelet transform based MDC and its application in image coding.