The measurement of color difference in machine vision is the basement of image segmentation based on the color features

but there is no colorimetrical system for machine vision. Traditional colorimetry has two major systems: (1) to build a color specification system experimentally and to use it as a reference (for example

the Munsell Book of Color). (2) to develop an analytical formulation based on discrimination experiments (such as

CIELAB

CIELUV). The color space based on the psychophysical experiments of human visual system is non-flat. That means the perceptual difference of colors can not be described by the Euclidian distances between the related two points in the color space and the map from perceptual difference to the Euclidian distances is not linear. In this paper

we built a virtual color space based on the physical models of light

reflectance of surfaces and the sampling function of visual system. In this virtual color space the chromatism of machine vision can be calculated linearly. We also suggest that the chromatism measurement of machine vision is different from the spectral reflectance of physical surfaces used in present computer models of color vision.