An Adaptive Intraframe DPCM Codec Based Upon Nonstationary Image Model

An Adaptive Intraframe DPCM Codec Based Upon Nonstationary Image Model By N. F. MAXEMCHUK and J. A. STULLER (Manuscript received November 21, 1978) This paper introduces a nonstationary model for images and develops an adaptive intrafield DPCM codec based upon the model. The codec attempts to minimize the mean-square coding error at each sample point in the picture. The quantizer in the resulting adaptive codec is found to be similar to that previously obtained from visual masking considerations. Comparative simulation results using 256 X 256pixel rasters are given for two- and three-bit/pixel versions of the adaptive codec, the three-bit/pixel Graham codec, and threebit/pixel previous element DPCM. I. INTRODUCTION This paper introduces a nonstationary model for images and develops an adaptive intrafield codec based upon the model. The codec adaptively estimates both the mean and probable range of values of the next picture sample to be encoded and adapts the predictor and quantizer accordingly. In so doing, the coder attempts to minimize the mean-square coding error (MMSE) at each sample point in the picture. The MMSE distortion measure is generally acknowledged to be a poor indicator of image quality.1 However, when it is applied on a point (rather than area) basis in conjunction with the image model presented here, the coder adaptation and resulting coding quality are found to be comparable to that previously obtained from visual masking considerations. This result follows from a property of human vision, stressed by Graham,2 concerning the strong connection that exists between image chaos (unpredictability) and the visual system's tolerance to noise-like coding distortion.