Russian manuscripts clustering based on the feature relation graph (FRG)

Abstract

Clustering of manuscripts becomes important nowadays because of the rapidly increasing number of documents in digital form. To solve this problem a new metric to compare handwritings based on the Feature Relation Graph (FRG) is investigated. This metric has demonstrated good results for the problem of text-independent writer recognition of Persian manuscripts on the basis of handwriting. Features that are based on local templates are extracted from manuscripts using Gabor and XGabor filters. We study the effectiveness of the most popular clustering algorithms for the problem of Russian manuscripts processing in the phase space of FRG. The paper presents numerical experiments demonstrating the effectiveness of the proposed metrics. The results of the various clustering algorithms are also provided.