Self-organization of high-order receptive fields in recognition of handprinted characters

The printed areas of a handprinted character with thick strokes were replaced by a frame formed by bended ellipses to represent the character efficiently and emulate high order receptive fields in a visual system. To afford topology preservation during adaptive matching of this frame with a template frame, we employ a devised self-organization model. This model uses these bended ellipses as training patterns in searching, measuring and updating their corresponding ellipses in the template frame. The neighborhood of a corresponding ellipse is also weighted by the appearance of the training bended-ellipse. With this method, each handprinted character can effectively evolve into its template character with predetermined training parameters. Each template has a different number of training cycles. Within this controlled number of cycles, the model can flex a handprinted character into a correct template.