Abstract
The standard model of contrast sensitivity is represented by the contrast sensitivity function (CSF) that shows sensitivity peaking around 4-8 cycles/degree and falling to zero around 40 cycle/deg. However, there are a number of problems with this model. One major problem is that it does a poor job at explaining our visual system's response to natural scenes that have amplitude spectra that fall off with frequency as 1/f. Although supra-threshold contrast matching studies show a flatter response function, this is still not a good match to natural scenes. In this talk, we introduce an approach to visual sensitivity that focuses on a neurons vector magnitude. We then propose the radical idea that visual sensitivity peaks around 30 cycles/deg. We provide a demonstration of this phenomenon with a chart of an array of log-Gabor functions plotting contrast magnitude against spatial frequency. We also show the results of a contrast matching experiment that supports this model. We argue that with this approach, the response to natural scenes is flat out to around 30 cycle/deg (i.e., the response effectively whitens the input). The approach explains why white noise appears to be dominated by high frequencies (not 4-8 cycles/deg) and argues that it is the highest frequency neurons that will produce the largest response to white noise (or a single pixel centered on their receptive field). We will note show how the optics of the eye modify these measures and show how eye movements may fit into this model (as proposed by Rucci and colleagues). Finally, we will use this approach to explain why the contrast sensitivity function has the shape that it does by distinguishing between the response magnitude of a neuron and the neuron's signal/noise level.