Quantization noise as a determinant for color thresholds in machine vision.

Color discrimination simulation is applied to study a uniformity of the color space of machine vision devices whose operation is based on a three-component color model and which involve analog-to-digital conversion of signals with a resolution of 8 bits per channel. Algorithms for finding the intervals of the dominant wavelength and color saturation of a specimen are developed. The spectral dependence of intervals of color parameters calculated using the digital images is found. It is shown that machine vision possesses the color discrimination thresholds, which can be drawn in the CIE1931 xy chromaticity diagram in the form of equal-contrast ellipses similar to the MacAdam ellipses. At resolution of 6 bits, the size of a reference MacAdam's ellipse is a little less than that of the machine vision ellipse's sizes, and at resolution of 7 bits, it is a little more than that of the machine vision ellipse's sizes. A hypothesis is proposed that implies that the process of an encoding of the visual neural signals may include procedures similar to an analog-to-digital conversion.