Chaotic CNN for image segmentation

A chaotic CNN associative memory that is able to perform complex pattern separation is presented in this paper. The introduced model has the form of a network composed of chaotic oscillators locally coupled by nonlinear conductances. A local pseudoinverse learning rule for binary pattern storage in the cellular memory structure is proposed. The chaotic units can temporarily synchronize or antisynchronize and hence retrieve patterns in the global synchronization state. Chaotic wandering from one pattern to another is an inherent property of the present model and allows object separation if a pattern superposition is presented to the network´s input