My research concerns the development and empirical testing of neural network models of visual processes, notably the segmentation, grouping, and contour formation processes of early and middle vision in primates. The modeling cycle begins with a conceptual analysis of core behavioral competencies evidenced in human psychophysical data. For example, consider the human capacity for long-range grouping of aligned contour fragments, such as in the perception of neon color spreading in the figure below..

A prototype neural design is sketched and, where possible, described mathematically and simulated on a computer. A computational model that addresses illusory contour formation and neon colors spreading includes, for example, analogs of cortical units with particular combinations of short-range and long-range connections and specific nonlinearities (rectification, saturation) in their signal functions. Computer simulation of model properties affords a more detailed comparison with physiological data than possible for less computationally specific formulations.

I look for occasions to further constrain model development by performing psychophysical experiments. For the case of illusory contour formation, Greg Lesher, then a graduate student, and I conducted a parametric study of the effects of line spacing and density on illusory contour formation, revealing a previously unreported inverted-U function for contour clarity as the number or density of inducers increases.

This study helped to guide parameter selection for balancing short-range lateral inhibition (competition) against long-range excitatory processes (cooperation) for contour completion in the model. Attempts to incorporate the implications of that experiment and other psychophysical and physiological data led, in turn, to a detailed local-circuit refinement of the original model. The latter version incorporates important recent results concerning the balance of disynaptic inhibition and monosynaptic excitation among pyramidal and spiny stellate cells of layer 3 of monkey V1, and shows how this circuit holds the key to long-range boundary completion.

While the example of illusory contour formation is illustrative, the content areas of my current research include textural segmentation, visual motion perception, and visual search. Please see the CNS Vision Lab page for a description of recent psychophysical work.