Authentication of Offline Signatures Based on Central Tendency of Features and Dynamic Time Warping Values Preserved for Genuine Cases

Abstract

This work proposes to authenticate offline signatures using a Case-Based Reasoner (CBR). The case base serves as a repository of sets of genuine signatures for which a central point on the n-dimensional global feature space is preserved along with the Inter-Quartile Range (IQR). These signatures are paired off to perform Dynamic Time Warping (DTW) comparison on their respective contours. Metrics generated from the global features and DTW values for the preserved signatures are utilized to predict authenticity of test signatures. Philosophically, CBR is a good classifier since it does not need any training by forgery models. The overall accuracy of the CBR classifier is maintained at a reasonably high value as a larger False Rejection Rate (FRR) is compensated by a tight False Acceptance Rate (FAR) value when compared with a MLP classifier. Both the classifiers have been tested on a standard offline signature database as well as one collected and prepared during the current research.