Chromatic Aberration, Linear Models, and Matching Color Images

near the accommodated wavelength can have detectable contrast in the retinal image, which implies that high spatial-frequency components play little role in color and contrast perception. Second, in the moderate spatial-frequency range, from 520 cpd, when the observer is accommodated to the yellow or green part of the spectrum, the visual system is dichromatic: there is no contrast in the short-wavelength receptor class. Perhaps most important, the OTF we calculated suggests an improved procedure for matching color images. The conventional method of setting point-by-point matches between images fails to account for the fact that image points on different displays may not have same pointspread function on the retina Since the spatial patterns on the retina from individual points on the displays do not match, one cannot match the retinal images of two points simply by adjusting the intensities of the three display primaries. Instead, to equate photo-pigment absorptions between images on different displays, one must adjust the primary intensities in corresponding spatial-frequency bands. (We describe this procedure in more detail below.) Because the OTF depends on the wavelength of the corneal image (as wed as its spatial frequency), using it to compute photoreceptor responses can be computationally quite expensive. When an image arises from a natural scene, representing the surface and illuminant spectral functions with finite-dimensional linear models greatly simplifies the computation. In that case, a simpler OTF can be computed that depends not on the wavelength of the corneal image but only on the weights of the basis functions that model the image. The number of weights win in most cases be much smaller than the number of wavelength samples, which is why the computation becomes so much less expensive. This simpler OTF can also be used to predict matches of color images on emissive displays. This is because emissive displays can be represented with a three-di-mensional linear model. We presented our algorithm for color matching on emissive displays in an earlier paper.7 Here we show that this use of the OTF is a special case of the OTF that arises from representing surface and illuminant functions with linear models. The remainder of the paper is organized as follows. In the next section, we introduce notation for the OTF and show how to use it to predict photoreceptor responses. Next, we introduce linear models for surface and illuminant spectral functions and show how their use simplifies the prediction of photoreceptor responses. Then we show how our Abstract