On feature extraction using gabor filter and feature relation graph for offline signature verification

Abstract

The most important and difficult stage of each offline signature verification system is feature extraction stage. The performance of the system mainly depends on effectiveness of the feature extraction algorithm. Current methods in this domain make use of different feature extraction and classifications approaches like Radon Transform, VQ, Gabor filter, SVM, KNN, MD, and etc. However, accuracy is still the main issue in this field. The final aim of this study is to implement an offline signature verification system to verify the originality of the test signature images and distinguish the skilled and random forgery from genuine. This project combines Gabor filter, XGabor filter, and gravity center point as a novel feature extraction algorithm and uses FRG (Feature Relation Graph) classifier for classification phase. The proposed system is validated using GDPS signature database, where it achieved equal error rate of 7.66% which is outperformed the latest works in this field.