Abstract: Combining ecological optics with computational vision, we claim that theunderlying task of color information processing is detecting light invariant,which is spectral reflectance of physical surfaces. On this foundation, we pro-pose a computational theory of color vision and a biological standard ofcomputational vision itself. Early color vigion is a typical ill-posed problem, some constraints mustbe introduced. The first two constraints are full-spectrum and full-color con-straint. After proof of the existance of color algorithm, a subjectivedescription of physical surfaces is constructed, that is relative reflectance. Itis reasonable that the relative refletance is a neural representation of color andbrightness perception. For the objectivity of color perception, the third con- straint, objective constraint, is proposed. As a result, the basic assumption forconstructing color perception is as follows. The early vision system constructs color perception in optic environment.If tke information processing fits well with the full -spectrum, full-color andobjectivity constraints, the constructed color perception is a neural representationof color and brightness. We expound the completeness of neural representation, the objectivity ofcolor perception, and some problems on performance of color vision. We arguethat brightness and color vision is unity and should be considered on the sameframework. Both differences of color and brightness in spatial-temporal domainplay an important role in conveying visual information. Some psychophysicaland physiological results are consistent to the computational theory of colorvision and the unity of color and brightness in early vision.

Key words: color vision, computational theory, light invariant, neural representation