Efficient Gaussian Source Coding Based on Distribution Mapping

Abstract

Gaussian sources are commonly found in the fields of mathematics and engineering.There is instantaneous code for geometric and two-sided geometric distributions; however,there is no simple,instantaneous code for normal distributions.Therefore,we have developed a method of mapping a normal distribution onto a geometric distribution before applying Golomb codes.Experimental results showed that our method consistently yielded coding efficiency of more than 98% for quantized Gaussian sources,which is comparable to the coding efficiency of second-order Huffman codes.For actual video coding residual signals,our codes were more efficient by six points than straightforward Golomb codes.Furthermore,we developed and demonstrated an accurate estimation method for determining optimal Golomb parameters,and we investigated the use of hybrid coding to enhance the coding performance.