Modeling of local neural networks of the visual cortex and applications to image processing

摘要

Abstract: A model of an iso-orientation domain in the visualcortex is developed. The iso-orientation domain isrepresented as a neural network with retinotopicallyorganized afferent inputs and anisotropic lateralinhibition formed by feedback connections viainhibitory interneurons. Temporal dynamics of neuronresponses to oriented stimuli is studied. The resultsof computer simulations are compared with those ofneurophysiological experiments. It is shown that thelater phase of a neuron response has a more sharporientation tuning than the initial one. It issuggested that the initial phase of a neuron responseencodes intensity parameters of visual stimuli, whereasthe later phase encodes its orientation. The design ofthe neural network preprocessor and the architecture ofthe system for visual information processing, based onthe idea of parallel-sequential processing, areproposed. An example of a test image processing ispresented.!