Double opponency as a generalized concept in texture segregation illustrated with stimuli defined by color, luminance, and orientation

We present a series of experiments in texture discrimination with textures whose local elements are defined by their color (red and/or green), luminance polarity, and orientation (vertical and/or horizontal). The 23 distinct texture configurations were designed for testing and parameterizing a model based on the generalization of the concept of double opponency. The double-opponent mechanisms of the model are classified into four domains: the luminance- and color-oriented (LO and CO) domains and the luminance- and color-nonoriented (LnO and CnO) domains. Texture edge strength is extracted from the weighted, half-wave rectified outputs of each double-opponent mechanism. These signals are then combined through probability summation within each domain. Overall sensitivity to a given texture pair is predicted by the probability summation of all the domain outputs. Good fits of the experimental data are obtained when the contribution of the CO domain is the smallest and the contribution of the CnO domain is the largest. We suggest that the double-opponency principle is a natural way of implementing spatial interactions in higher-order domains and that it could be generalized to other dimensions such as spatial frequency, motion, and stereopsis.