A Dynamic Pattern Recognition Method Using the Preview Controlled Saccadic Movement of the Detector

The purpose of this paper is to suggest the use of biological movements of detectors for pattern recognition. The authors demonstrate a unique method for recognizing patterns composed of straight and curved lines such as characters or edge patterns, independent of position, rotation, size, or slight deformation. The principle of the method is, different from conventional scanning techniques, based on preview controlled curve tracking movement similar to the saccadic motion of the eye. The recognition depends on the time sequence of straight line, curve, angle, intersection, etc. Discrimination between a straight and a curved line, larger or small curvature, and obtuse or acute angle is based on the relative difficulty in tracking the line. For small patterns which lie completely within the field of the detector, a well-known multi-layered recognition device is superior. For larger patterns, however, which lie partially outside the field, the method proposed here is superior. The authors think that their method could be useful for the eyes of robots when combined with the kind of multi-layered pattern recognition devices described by Fukushima.