Modular neural networks for shape and/or location recognition

The ability of the brain for visual pattern recognition is quite flexible. The present paper proposes a neural network model that simulates this flexibility of visual pattern recognition. The model is an integration of two independent modules; one called Shape Recognition Module (SRM) and the other named Location Recognition Module (LRM), whose functions are directly suggestive from their names. The integrated model was developed as a visual recognition module for a more complicated neural network for discovering regularities in alphabetical letter series (Anzai, Mori, Ito and Hayashi, 1987). To find regularities in a given series of characters, the shape and/or serial location of each letter must be encoded to construct some internal representation. The model described in this paper takes care of this encoding process.