A Study of Pattern Recognition by the Use of Walsh Orthogonal System

A method for pattern recognition by the use of Walsh orthogonal system has been studied by many scholars. However, these methods for pattern recognition possess two weak points as follows. (1) A high recognized percentage can not be obtained when input patterns have distortinons. (2) Input patterns become high dimensional pattern vectors and a system for recognition become large. For these reasons, first a quantized pattern is mapped into a low dimensional one to get rid of two weak points in the persent method. On account of this result, processing-time and memory capacity cut down in the use of a digital computer. Secondly, the variation of the recognized percentage is examined against the reduction of dimension in Walsh orthogonal space from a point of view that the fewer number of feature parameters makes the more symple system. Finally, a computer-simulated experiment is carried out by the method described above and the results of the experiment are shown.