片桐滋
A UNIFIED APPROACH TO PATTERN RECOGNITION
Abstract:Pattern recognition is a complex process consisting of several subprocesses, such as
Feature extraction and classification. Naturally, design efforts should be devoted to the
entire process of recognition in a manner consistent with task goal achievement, i.e.,
achievement of high recognition accuracy. Recent approaches based on Artificial Neural
Networks (ANNs) have contributed to increasing recognition accuracy. However, they are still
insufficient due to the lack of a global scope for the design of an overall recognition
system. This paper reviews the present situation of the ANN-based approach from the
methodological viewpoint of the Generalized Probabilistic Descent method and clarifies a
number of technical issues to be further investigated.