Importance

Loading Importance Indicators...

Abstract

In a feature extracting device for pattern recognition which is hardly affected by a pattern variation, an input pattern received by a learning pattern input/store unit is normalized by a normalizing unit and projected on a subspace group by a feature vector extracting unit, so to calculate feature vectors from each projection length, and a parameter updating unit updates basis vectors of each subspace so as to increase the ratio (variation between classes/variation within a class) as for the feature vectors, and optimizes them in a way of absorbing the pattern variation into each subspace according as the learning by the update processing progresses, thereby realizing the high ratio of the variation between classes to the variation within a class at the time of completing the learning and enabling the feature extraction more suitable for pattern recognition.

Loading the Abstract Image...

First Claim