Quantization of TV Chrominance Signals Considering the Visibility of Small Color Differences

A quantizing method for chrominance signals is presented which takes into account that the visibility of small color differences and the areas of reproducible colors are a function of luminance. For the evaluation of quantizing characteristics and for an estimate of the effectiveness of such a procedure, CIE (Commission International de l'Eclairage) formulas are used. It is proposed to subdivide the color space into several layers of constant luminance signal and to choose, switched by the luminance signal, an individual quantizer for each layer. Such quantizers are calculated for two different coding schemes: quantizing the TV chrominance signals
U_{R-Y}
and
U_{B-Y}
directly or quantizing chrominance signals which have been transformed into a color space derived from a neurophysiological model of vision. Because of practical reasons the quantization of
U_{R-Y}
and
U_{B-Y}
is preferred and elaborated in more detail. The calculated non-linear quantizing grids are approximated by two different methods, with small rectangular elements, or they are piecewise linear approximated by interpolation. To achieve a color difference of
\DeltaE_{1964 CIE} = 4.4
as maximum distortion, a PCM quantization of 5 bits in
U_{R-Y}
and
U_{B-Y}
is required. First order entropies indicate an average bit rate of 3.5 for
U_{R-Y}
and 2.7 for
U_{B-Y}
.