Object recognition using neocognitron

The implementation of a neural network model called Neocognitron is presented in this thesis. The Neocognitron is a hierarchical, multiresolution, self-organizing structure (learning without a teacher) that recognizes objects on the basis of the geometrical similarity of their shapes. First, the basic structure of the Neocognitron was modified for the recognition of multiple inputs. Next, the Neocognitron was expanded into a modular design in order to acquire the ability to recognize a large number of classes. The modular design consists of a number of Neocognitron-type networks that are connected in parallel and that receive input from a single input layer. The modular version of the Neocognitron was trained with five different groups of alphanumeric characters.