Writer identification system for Ethiopic handwriting

Abstract

Writer identification is a popular and ongoing research area having a wide variety of applications in banking, criminal justice system, access control, determining the authenticity of handwritten mails, etc.

In this paper, an off-line text independent Ethiopic writer identification system has been proposed. The system uses 50 handwritten text blocks collected from 25 volunteers (each person was made to write on two A-4 size pages in Ethiopic handwriting). These text blocks are scanned and stored for further processing by the identification system.

Two approaches have been employed for feature extraction from the handwritten images: texture level using multi-channel Gabor Energy Features and the character-shape (allograph) level using codebook of connected component contour.

Experimental results demonstrate that 93% correct identification, in a hit list of size 3, and 76%, in a hit list of size 1, using Gabor energy features and 96% correct identification, in hit list of 3, and 92%, in hit list 1, using codebook of connected component contours, are acquired.

Keywords: writer identification, writer verification, image pre-processing, multi-channel Gabor filtering, Gabor Energy Feature, Connected Component Contour and Codebook.