Neural network paradigm for visual pattern recognition

A neural network for visual pattern recognition is proposed and has been successfully applied to the task of handwritten character recognition. The same network can also be used for shape identification and other 2-D visual pattern recognition tasks. The neural network performs two functions; feature extraction and pattern classification. The feature extraction layer identifies the dominant geometric features of the preprocessed image. Once the features have been extracted, a second layer maps the feature vectors to a lower dimension feature space, and third layer maps the respective points, in the reduced feature space, to corresponding points in the classification space. The network is trained using a combination of a self-organizing algorithm, for the feature extraction layer, and supervised training, for the classification stage.<
>