Pattern recognition by combined invariants

A feature-based recognition of objects or patterns independent of their position, size, orientation and other variations has been the goal of much recent research. The existing approaches to invariant two-dimensional pattern recognition are useless when pattern is blurred. In this paper, we present a novel pattern recognition system which can solve the problem by using combined invariants as image features. The classification technique we choose for our system is weighted normalized cross correlation. The mean of the intraclass standard deviations of the kth feature over the total number of prototypes for each class is used as a weighting factor during the classification process to improve recognition accuracy. The feasibility of our pattern recognition system and the invariance of the combined features with respect to translation, scaling, rotation and blurring are approved by numerical experiments on head images.