A theoretical model for layered visual processing.

A theoretical general model of layered computation in the retina is presented. Each layer is functional in the sense that it may correspond or not to an anatomical layer. It is formed by computing elements which can perform in principle any non-linear arbitrary function on a three-dimensional input space. This space consists of 2 spatial dimensions, plus time. The function performed by each computing element of a layer is simplified for the cases of invariance, space or time linearity and for time independence. Two illustrations are also presented. The first is a model of the simple ganglion cells in cat's retina. The second, a model of the group 2 ganglion cell of the frog's retina.