STATISTICS FOR EFFICIENT LINEAR AND NON-LINEAR PICTURE ENCODING

Abstract

Differential pulse code modulation (DPCM) of video signals can be improved by assuming the model of the general Markov Source, which has not only linear but also nonlinear statistical dependencies. The proposed type of coder will have a fixed code word length and synchronous bit-rate. It consists of a differential coder with linear prediction and a quantizer with an additional nonlinear coder. The nonlinear coder can be considered as changing the quantizing characteristic according to the past of the signal. The improvement over PCM and over DPCM is measured by the signal to quantizing noise ratio. Picture material under investigation consists of 4 different single frames of Standard CCIR 625- line television format with relatively high detail. Two types of linear prediction have been investigated : one dimensional (previous element) and two dimensional spatial prediction. The nonlinear coder is controlled by up to three previous elements of the same and the previous line. The measurements show an improvement over previous element DPCM of up to 9 dB S/N which would result in additional saving of 1.5 bits per coded element.