Image recovery from noisy transmission using soft bits and Markov random field models

A new method for robust image recovery from noisy transmission based on soft bits and Markov random field modeling is proposed. The method aims to exploit the residual redundancy present in the symbols produced by a source encoder together with knowledge about the statistical properties of natural images. The soft-bit information required for reconstruction can be extracted either directly at the output of a noisy communication channel or at the output of a soft channel decoder (if channel codes are used). In the latter case, the proposed algorithm can also be used in conjunction with iterative channel decoding, resulting in joint source-channel turbo decoding. Examples are presented for the case where no channel codes are employed. The results obtained indicate that the proposed method yields very good performance even under extremely noisy conditions.