Off-line persian signature verification: An empirical evaluation

Abstract

Signature verification is a frequent task in forensic document investigation. In this paper, an empirical study of off-line Persian signature verification system has been proposed. Persian signatures are different from other signature types because citizens generally do not use text in it and they draw a outline as their signature. Initially scanned signature images undergoes suitable preprocessing steps. After preprocessing, features based on Euler number, average object area, mean and area are extracted. Finally offline signatures are verified using SVM, KNN and Boosted Tree. Publically available database UTSIG is used. Paper also compares the result for different signers and different samples size.