Modeling shape representation in visual cortex area V4

Abstract

Visual processing in biological systems is classically described as a hierarchy of increasingly sophisticated representations, originating in primary visual cortex (V1), progressing through intermediate area V4, and ascending to inferotemporal cortex. The computational mechanisms that produce representations in intermediate area V4 have remained a mystery. In this thesis I show that the standard model, a quantitative model which extends the classical description of visual processing, provides a computational mechanism capable of reproducing and predicting the responses of neurons in area V4 with a translation invariant combination of V1 responses. Using techniques I have developed, model neurons accurately predict the responses of 8 V4 neurons to within-class stimuli, such as closed contours and gratings, and achieve an average correlation coefficient of 0.77 between predicted responses and measured V4 responses. Furthermore, model neurons fit to a V4 neuron's grating stimulus response, can qualitatively predict the V4 neuron's 2-spot reverse correlation map. These results successfully demonstrate the first attempt to bridge V1 and V4 experimental data, by describing how representation in V4 could emerge from the nonlinear combination of V1 neural responses.