Structural feature based approach for script identification from printed Indian document

Script identification is a complex real life problem for automation of printed or handwritten document processing. The task becomes more challenging when it comes about a multi script/lingual country like India. For the development of OCR for a particular language the script needs to be identified first. That is why development of a script identification system is a pressing need. Till date no such work is available considering all 13 official Indian scripts. In this paper we present a scheme for script identification from printed document for 10 official Indian scripts namely Bangla, Devnagari, Roman, Oriya, Urdu, Gujarati, Telegu, Kannada, Malayalam and Kashmiri. Total 459 document pages are considered and 62 dimensional feature set is computed for the present work. Finally using simple logistic classifier with 5 fold cross validation an average identification rate of 98.9% is found.