Abstract

A model of signal detection based on an idealized neural contrast detector is described. The neural mechanism spatially sums photons according to a receptive field weighting function that subtracts out the dc stimulus level. The statistics of photon absorptions are taken into account, so that the time-integrated receptive field output is treated as a random variable. In the most general case, this variable is power law transformed. When the resulting variable exceeds a fixed neural threshold, a neural spike is generated. We investigate the performance capabilities of an ideal observer who monitors the spike train of a single neural mechanism which is ideally scaled and retinally located for detecting a flash of luminance ΔL, area A, and duration T, superimposed on an adapting field of luminance L. At any stage in this process, the noise in the system can be multiplied and/or added to by a constant amount of internal noise, without loss of generality. For large L, most of the variability in the spike interarrival times is a result of photon noise. For small L, internal noise dominates. Further assuming judgments based on a fixed number of spikes, the model correctly accounts for the shapes of TVI curves—including absolute threshold, a deVries-Rose region, an asymptotic Weber region—and their dependence on flash area and duration. Many properties of the time course of visual response are also accounted for. These include data concerning visual reaction time, persistence, temporal adaptation, and brightness transients.

© 1991 Optical Society of America