Methods for Designing Differential Quantizers Based on Subjective Evaluations of Edge Busyness

Experience with Differential Quantizers 1,2,3 teaches us that noise superimposed on brightness boundaries of a picture is less objectionable than similar noise on areas of uniform brightness. This work is an attempt to evaluate the phenomenon and make efficient use of it when designing quantizing scales for Differential Coders. We know that there is advantage in companding the quantization levels so that the largest steps are used for reconstructing rapidly changing signals and the smallest steps for slowly changing signals. In the past, the companding laws for differential quantizers have usually been obtained experimentally by trial and error, but these techniques are inadequate when there are a large number of levels. A more objective method is described here; it provides a basis for theoretical synthesis and evaluation of companding laws independent of the number of levels that are required. The essence of the method is a subjective evaluation of the dependence of the visibility of noise on the rate of change of the signal. The measurement is used to determine how the visibility of the net quantization noise in a picture depends on the companding law. Laws that minimize 1495