Finding arbitrary shaped clusters for character recognition

Abstract

Several supervised and unsupervised methods have been applied to the field of character recognition. In this research we focus on the unsupervised methods used to group similar characters together. Instead of using the traditional clustering algorithms, which are mainly restricted to globular-shaped clusters, we use an efficient distance based clustering that identifies the natural shapes of clusters according to their densities. Thus, in the case of character recognition, where it is natural to have different writing styles for the same character, the algorithm can be used to discover the continuity between character feature vectors, which cannot be discovered by traditional algorithms. This paper |introduces the use of an algorithm that efficiently finds arbitrary-shaped clusters of characters, and compares it to related algorithms. Two character recognition data sets are used to illustrate the efficiency of the suggested algorithm. © 2008 Springer-Verlag Berlin Heidelberg.