Abstract
We investigate a nonlinear two-stage network optimized to reduce statistical dependencies in natural images.This network serves as a model for the neural information processing in the higher visual areas of primates (visual cortices V2,V4). It is analyzed with regard to nonlinear selectivity and invariance properties. We show that the proposed network principle leads to units that are highly selective with respect to the input signal space and to units that are invariant with respect to certain stimulus classes. A special property of the system is the emergence of nonlinear frequency interactions, which are necessary for exploiting the higher-order statistical structure of natural images. We extend the concept to multi-layer systems and present some simulation results.
Supported by DFG (SFB TR 6023) and BMBF (BCCN Munich 01GQ0440)