Predictive image coding using multiresolution multiplicative autoregressive models

The authors present a new image coding technique called multiresolution multiplicative autoregressive coding. It is a hybrid method that integrates multiresolution interpolative coding and two-dimensional predictive coding based on multiplicative autoregressive models. In addition, the proposed scheme utilises newly developed methods for block adaptive quantisation of prediction errors and fixed-to-variable-length block coding. The performance of the proposed scheme is tested on three different images and, in all cases, reconstructed images of reasonably good quality are obtained at bit rates less than 0.35 bits per pixel. The bit rates and the signal-to-noise-ratios of the reconstructed images compare favourably with the existing low bit rate image coding methods.<>