Abstract

We claim that color vision provides important information about properties of surfaces. To process the chromatic information in images, one has to use a model of color vision and determine its responses to different types of input. Our model consists of different operators which resemble in their characteristics the different layers in the biological chromatic visual pathways. In particular, we look at color-opponent and double-opponent cells with center-surround spatial organization. We characterize the mathematical properties of these units and test their responses with respect to changes in wavelength intensity and space. Our results¹ show that the opponent operators give information about the chromatic content of the input in terms of two mutually exclusive color pairs. Moreover, these operators distinguish areas which are isoluminant from areas which have different luminance levels. The double-opponent units detect color borders, no matter the luminance levels. They have bandpass properties, similar to the difference-of-Gaussians (DOG) operator,² with the advantage of responding to some stimuli which the DOG will not respond to. In addition, the double-opponent operators seem to disregard certain types of shadow, which leads to speculation about their inability to discriminate between the lit and shadow areas. This information, in coordination with responses from opponent operators, can be useful in detecting changes in materials.

© 1985 Optical Society of America