ADAPTIVE CODING USING TWO-DIMENSIONAL PREDICTION FOR IMAGE SIGNALS.

The major problems in predictive coding for image signals are granular noise occurring in flat parts of the intensity variation and slopeoverload noise occurring at outlines. With the aim of reducing these noises, a new coding system is proposed in this paper in which the quantizing level of the difference signal is varied by two-dimensional prediction of the properties in the neighborhood of the picture element to be coded, utilizing already-coded picture elements. The principle of this system is described, and responses at outlines of the intensity variation and the effects of transmission errors are studied. This system can be considered to utilize in an efficient way the information obtained from a two-dimensional prediction. The variable quantizing level for this purpose can be determined on the basis of the already-measured conditional distribution of the difference signal. Improvement of the quantization noise in the proposed system is shown by computer simulation in comparison with the conventional system.