Abstract
Simultaneous contrast and White's effect, as predicted by current models of brightness coding (Dakin and Bex, 2003; Blakeslee and McCourt, 1999), can be shown to be reliant on the non-linear amplification of low spatial frequency information, sampled continuously by large receptive fields (< 0.5cpd). However, the existence of receptive fields of this scale in V1 is controversial. Our goal is to predict both simultaneous contrast and White's effect only using filters of the scale typically found in V1. A model is presented, which is based on the proposition that V1 simple cells provide a measure of the 1st and 2nd order derivatives of a blurred version of the retinal image, and that this information is sampled at discrete intervals by overlapping receptive fields. An approximation of the Taylor series representation of the image, operating on the available differential information, is used to fill-in brightness in the regions local to each sampling point. Reconstruction of the 0th order term (the global brightness structure), is performed by enforcing continuity between neighboring regions of local brightness. This is achieved by shifting the absolute brightness level of each local region, and optimizing the shifts such that the variance in the differences in brightness within overlapping regions is minimized. This technique can effectively reconstruct natural scene images. Error in reconstruction of the global brightness structure by this method provides an account of simultaneous contrast and White's effect.