Abstract
Background: A computational model that is capable of explaining the effects of contrast adaptation when testing at threshold contrast and at supra-threshold contrast levels is proposed. The model assumes that the effects of adaptation may be both orientation specific and isotropic. Methods: The model codes for image contrast by the output of spatiotemporal frequency tuned channels whose individual responses are normalized by a summation of individual channel responses present in a neuronal pool (Heeger, Vis. Neurosci., 1992). Each spatiotemporal channel is assumed to have an additive and multiplicative noise component: the multiplicative component arising from an uncertainty in the individual channel weights. With this noise model, the orientation specific component of adaptation is explained by a subtractive de-noising strategy that leads to the familiar rightward shift of the linear filter's contrast response function. The isotropic component of adaptation is represented in the model by a narrowing of the spatial bandwidth and a broadening of the temporal bandwidth of the individual channel weights. Results: The model can explain: (i) why the orientation specific elevation in threshold contrast as a function of adaptor contrast can saturate when the adaptor approaches 10% contrast (Snowden, JOSA, 1994); and (ii) why the isotropic components of adaptation may be described by the two components of signal amplification and division (Ross and Speed, Vis. Res. 1996). Conclusion: The proposed model suggests that the process of contrast adaptation may be viewed as an optimal de-noising strategy. The low level of adaptor contrast at which the elevations in threshold contrast can saturate appears to be inconsistent with the view that the orientation specific effect of adaptation represents a strategy used by the visual system to combat the problem of neural saturation in neuronal response firing rates.