A Neural Network Model Sensitive To Oriented Slabs

Visual pattern recognition aspects emerging as collective properties of systems of neurons are considered. In this sense, the firing activities of groups of individual neurons are seen as the elementary entities for cooperative phenomena processing. Specifically we show as massively coupled neural assemblies with antisymmetrical synaptic junctions can exhibit orientational selective properties according to the behaviour of simple cells located into mammalian V1 area. Biological and theoretical supports suggest that information is represented in the nervous system by a small number of highly connected neurons. In this paper a neural network approach to emulate orientational sensitive simple cells behaviour is taken into account. It is based on assemblies of 32 elements like neurons arranged in an isotropic mode with antisymmetrical synaptic patterns. The dynamicisms of the system are described by a dynamical linear model producing particular oscillating trajectories in the state space. The resulting system is trained to recognize specific orientation by a Hebb-like rule reinforcing the synaptic strengths activated by the input stimulus. Experimental results using this approach are presented that will respond to test patterns of related inputs.