Individuality of Isolated Bangla Characters

Abstract

Writer Identification and Verification, a behavioral bio-metric study has gained a renewed interest in research work for its promising prospect in real life applications. In current times, to the best of our knowledge there is no complete system of Writer Identification / Verification on Indic scripts including Bangla. In this paper a scheme has been proposed for individuality of handwritten and writer identification on Bangla script. The scheme has prospect not only in Writer Identification but also in Writer Verification, Graphological Analysis and in various field of Handwriting Forensic. As there is no such standard Bangla writer database, for the proposed system a database consisting of total 31950 characters and vowel modifiers from 90 writers with 5 sets from each writer has been developed. Standard and robust features like 64 and 400 dimensional has been used for the evaluation of the quality of the collected database. The LIBLINEAR and Multilayer Perceptron classifiers of WEKA tool has been used for analysis of the characters and vowel modifiers. An observation of each characters and vowel modifiers reveals that the character GA ("/") has the highest individuality of 55.85% and modifier REF ("/") has lowest individuality of 11.05% for MLP classifier on 400 dimensional feature. The individuality has been calculated by finding out the writer identification accuracy for individual character. Writer identification accuracy of 99.75% has been achieved for 90 writers with 5-fold cross validation.