Pattern recognition using hierarchical feature type and location

In the human vision, the feature detection based on the line are hierarchically processed. In addition, they are separated into two parts: one is the feature type, and the other is the feature location. In this paper, a new model of pattern recognition using the hierarchical feature types and their location is proposed and realized by using the multilayered neural network. Line features are detected as lower feature. Then, more complex features, based on how line features are crossed, are detected as higher features. These higher features are processed in both their types and location. Also, training and pre-recognition are separately processed. Total recognition is performed by using these results. The model has a feedback signal in the feature detection block, so that it can control the feature detection process. Computer simulation of character recognition shows the effectiveness of the proposed model.