Subjectively Optimal Quantization of Pictures

The problem of quantization has been thoroughly studied for random processes. The usual criterion is to minimize the mean square error. Because of the eye's characteristics, this does not lead to the least number of bits. In this concise paper the results of experiments on gray scales and on quantized pictures are presented which lead to the choice of a compressor with a subjectively optimal parameter in the class of logarithmic compressors. Comparisons are made between the minimum mean square error quantizers on luminance signals, the usual quantizer for television signals (linear quantizer after a gamma correction) and the proposed quantization scheme designed for real visual scenes.