Prototype setting for elastic matching-based image pattern recognition

The purpose of this paper is to emphasize the importance of the consistency between the distance measures on prototype setting and discrimination in elastic matching (EM)-based recognition. Specifically, this paper focuses on the following points: (i) confirmation of performance degradation when Euclidean distance is used on prototype setting whereas EM-distance is used on discrimination, and (ii) proposal of new prototype setting algorithm where this inconsistency is avoided. Through an experiment of handwritten character recognition, the effectiveness of the proposed algorithm was quantified.