Appearance and detection of colored patterns
I present and test a model for the appearance and detection of colored patterns. I refer to the two part model as the separable-independent model. First, three linear mechanisms encode the color properties in the image. Second, the sensitivity of each mechanism is independently scaled according to the spatial pattern. The probability of correct detection is monotonically related to the vector length of the three channel outputs. For appearance, locations in two stimuli with different spatial patterns will have the same color appearance when the output from the three channels is the same.
To test the model for appearance, I measured the degradation of color perception with increasing spatial frequency using an asymmetric color-matching task. Subjects adjusted the color appearance of a two degree, foveal, uniform square to match the color appearance of one bar in a squarewave grating modulated about a neutral gray field. Subjects made complete color matches to spatial frequency patterns ranging from one to eight cycles per degree (cpd) and nine color directions for each frequency squarewave. For spatial patterns beyond eight or ten cpd, there is hardly any color (or even gray-scale) vision altogether. I report data from two subjects who each made 750 appearance judgments. To test the model for detection, I measured threshold performance using a two-interval forced-choice procedure. Subjects collected threshold data to several different spatial frequency Gabor patches (up through eight cpd) in many color directions. I summarize about 75,000 forced choice decisions.
I have compared the separable model with several less restrictive models that do not require separability. I see no reason to reject the model for either experimental data set. The fit to the separable model defines three mechanisms with separable spatial and spectral sensitivities. Data from both experiments and all subjects are explained by two spectrally opponent, spatially lowpass mechanisms and one all positive color mechanism. For the appearance data, the positive color mechanism is spatially bandpass for one subject and lowpass for the other. For the detection data, the positive color mechanism is spatially bandpass.