A quantitative model relating visual neuronal activity to psychophysical thresholds

Abstract

We investigate how a simple, physiologically motivated three-stage neuronal model can establish a quantitative relationship between activities in small populations of simulated early visual neurons and human psychophysical thresholds. The model consists of: First, a bank of linear filters tuned for orientation and spatial period; second, non-linear interactions between filters; and, third, a statistically efficient decision stage. The model quantitatively reproduces human thresholds for five classical pattern discrimination tasks, using a unique set of automatically determined parameters. The resulting model components are all plausible in terms of putative neuronal correlates.