Feature extraction for a multiple pattern classification neural network system

Feature extraction is an essential problem in pattern classification. The success of a pattern classifier very much depends on the effectiveness of the features representing the patterns of different classes. In multiple pattern classes, it is important to find features that can be used to discriminate each class from all the other classes. This paper presents an algorithm for feature extraction from a training data set followed by a neural network system for multiple pattern classification. We have applied the system to two different applications, handwritten digit recognition and occupant classification. The results show that the proposed feature extraction algorithm is a promising technique.