Figure 4
From: Evidence for the intrinsically nonlinear nature of receptive fields in vision
A simple L+NL model, consisting of a difference of Gaussians (DoG) linear RF followed by a pointwise nonlinearity, has to change with the input in order to reproduce the crispening phenomenon. When the model is adjusted to the uniform background condition (red curve), it qualitatively replicates brightness perception for the uniform background case (blue curve, left) but not for the salt and pepper surround case (blue curve, right). The reverse happens when the model is adjusted for the salt and pepper background condition. Both the DoG filter and the nonlinearity change with the stimuli (see “Methods”).
