J. Andrew Henrie

Population models for coding orientation in the visual cortex, and the effect of stimulus size

New York University

We have previously analyzed the classical psychophysical relationship between orientation acuity and stimulus size by applying an optimal decision model to the output of a population of physiologically inspired linear spatial filters (Henrie & Shapley, 2001). The implementation of this model failed to account for the relatively slow falloff in orientation acuity with decreasing stimulus size. This model estimates orientation acuity by application of optimal estimation techniques to the half-rectified response of a bank of linear spatial operators spanning all orientations whose size and form (Gabor) were matched to the single unit properties of primary visual cortex (V1) neurons. This model, which has been applied to many visual psychophysical phenomena, predicted a much faster decrease in acuity with decreasing stimulus size than observed behaviorally. The suggested reason for this discrepancy is a fundamental inadequacy of a static linear filter at capturing the response selectivity of V1 units. In the work I will talk about here, the size of the linear Gabor filter (the first stage of the model) was decreased until the model's output fit the behavioral relationship between orientation acuity and stimulus size. The spatial tuning properties of these relatively tiny Gabor filters was then compared to the tuning properties of V1 neurons. There is a considerable discrepancy between these fitted Gabor functions and measured V1 receptive fields. We interpret the discrepancy between the linear filter which can account for the data and the observable form of V1 selectivity as a reflection of a non-linear interaction among the individual elements in V1. In particular, non-linear suppression (e.g. inhibition) from units preferring different stimuli may be able to explain this. For example, divisive inhibition has recently been used as a single explanation for detection and discrimination of spatial stimuli (Itti et al. 2000), and Ringach et al. (submitted for publication) found physiological evidence that highly selective neurons in V1 require non-linear suppression to account for their response.