A Lexicon Reduction Method Based on Clustering Word Images inOffline Farsi Handwritten Word Recognition Systems

In this paper a novel approach for lexiconreduction of Farsi words is proposed. For this purpose weextract upper and lower profiles, vertical projection profileand black/white transition from word images. Using DTWsimilarity between words in the database is measured.The Isoclus algorithm is used to cluster handwritten wordimages of training dataset. The initial center of clusters isdetermined from agglomerative hierarchical clusteringalgorithm.Experimental results on IRANSHAHR dataset show apromising result. It yields a lexicon reduction of 77% withaccuracy of 94%. We also evaluate the proposed systemwhen combination of statistical features and different typeof distance measures are used.

کلیدواژه ها:

Farsi Handwritten recognition ، Isoclus algorithm ، agglomerative hierarchical clustering ، dynamic time warping andwavelet transform