Offline Handwritten Signature Verification System

Abstract

Even before the era of computers, handwritten signature was being used as a unique biometric. There are two methods that extensively deliberated the signature verification. They are Offline method and Online method. Even if it is considered somewhat difficult than online method due to the need of dynamic information, offline systems are straightforward to make use of when compared to online systems. Because of its importance for use in day-to-day life, the offline verification system has taken more attraction. This document presents an offline handwritten signature verification system using the support vector machine approach. Features are extracted from the signature images by calculating, Grey level Cooccurrences Matrices (GLCM). Then the texture feature calculations are performed and the SVM model gets trained with them. The appropriate SVM parameters (Gamma and C) are obtained by performing a k-Fold Cross-Validation by trying out different parameter combinations. In the verification process the texture feature calculation of the disputed signature image is performed and obtained the feature vector to be verified with the trained SVM Model. And finally, in the classification, verification result is classified as genuine or forged. This method takes care of skilled forgeries. The main objective of the solution is to minimize the two important parameters False Acceptance Rate (FAR) and False Rejection Rate (FRR) usually used in any kind of signature verification system. The proposed system has achieved a performance of approximately 86% by using a dataset of 768 signatures (genuine signatures and skilled forgeries) from 32 writers.