A neural network architectural model of visual cortical cells for texture segregation

A three-layer hierarchical neural network architecture to be used in early vision processing tasks (e.g., texture segregation) is presented. Taking into account both the linear properties of simple cells receptive fields and the nonlinear properties of intracortical processing, the structure and the functionality of simple, complex and hypercomplex cells are defined. The introduction in the model of hypercomplex cells, which interact with complex cells, provides a complete feature extraction of textured images. Specifically, the first layer of the network extracts oriented textured elements, the second layer increases the sensitivity to texture differences, and the last layer improves the selectivity of textural elements on the basis of their size.< >