Use of ARMA predictors in the differential encoding of images

Differential pulse code modulation (DPCM) has long been a popular method of data compression for speech. It has not been as popular for images, partly because of its poor edge performance. This is true even for adaptive DPCM (ADPCM). A paper by J.D. Gibson (IEEE Trans. Acoust., Speech, Signal Processing, vol.ASSP-33, p.1166 (1985)) provides a clue to the reason for this shortcoming. His analysis is used to motivate the design of backward adaptive ARMA (autoregressive moving-average) predictors for use in differential encoding of images. Simulations using these predictors validate the approach as these systems show significant objective gains as well as noticeable subjective improvements in the edge performance. An added advantage to using this approach is the improvement in noisy-channel performance.>