Individuality of Handwriting

Advisor: Prof. B. B. Chaudhuri

In this project I developed a tool which can classify different handwriting samples based on their authors and this was implemented for the Bengali script which is a derivative of the Devanagari script. I experimented with different pattern classification algorithms including feature representation and stroke-based Hidded Markov Model classification schemes. We found out that GSC (gradient, structural and concavity) features can be used to represent every character of the Bengali alphabet with 512 bits where each bit represents a feature (stroke-based). This model not only enabled us to recognize authors from handwritten manuscripts but also could be used to identify characters written by the same author upto a high degree of accuracy (99.3%).

(Detailed Report) (Certificate)