Joint source and channel decoding of compressed still images

Abstract

A joint source-channel decoding approach is described, which utilizes the residual redundancy in the bit stream of a source-coded image by employing a modified APRI-SOVA (a priori soft-output Viterbi algorithm) at the channel decoder. The source coders used are the nonadaptive discrete cosine transform (DCT) coding and subband coding with Lloyd-Max quantization. The log-likelihood ratio describing the a priori information is linearized. The reliability of the current bit is computed as a proper linear combination of the soft values characterizing statistical dependency of the zeroth order (probability of 1's or 0's) and the first order (bit correlation among neighboring samples). Substantial subjective and objective performance gains can be obtained by using this method. It is demonstrated that especially in a noisy environment, it may be better to keep the residual redundancy in the data and use it at the receiver side in order to achieve a more reliable decoding, rather than to further remove it by using a variable length coding. The trade-off between the allocation of bits for source and channel coding is also addressed.