NEURAL MODEL OF VISUAL STEREOMATCHING - SLANT, TRANSPARENCY AND CLOUDS

Stereomatching of oblique and transparent surfaces is described using a model of cortical binocular 'tuned' neurons selective for disparitie s of individual visual features and neurons selective for the position , depth and 3D orientation of local surface patches. The model is base d on a simple set of learning rules. in the model, monocular neurons p roject excitatory connection pathways to binocular neurons at appropri ate disparities. Binocular neurons project excitatory connection pathw ays to appropriately tuned 'surface patch' neurons. The surface patch neurons project reciprocal excitatory connection pathways to the binoc ular neurons. Anisotropic intralayer inhibitory connection pathways pr oject between neurons with overlapping receptive fields. The model's r esponses to simulated stereo image pairs depicting a variety of obliqu e surfaces and transparently overlaid surfaces are presented. Far all the surfaces, the model (i) assigns disparity matches and surface patc h representations based on global surface coherence and uniqueness, (i i) permits coactivation of neurons representing multiple disparities w ithin the same image location, (iii) represents oblique slanted and ti lted surfaces directly, rather than approximating them with a series o f frontoparallel steps, (iv) assigns disparities to a cloud of points at random depths, like human observers and unlike Prazdny's (1985) met hod, and (v) causes globally consistent matches to override greedy loc al matches. The model represents transparency, unlike the model of Mar r and Poggio (1976), and it assigns unique disparities, unlike the mod el of Prazdny.