User Authentication using On-Line Signature and Fingerprint Biometrics

Publications

Sumit Chachra, Himanshu Sahani, Vinod Kumar and H.K. Verma, ‘Adaptive Segmentation of On-Line Signatures’, Accepted for presentation at the 11^th International Graphonomics Society Conference, Scottsdale, USA, 3-5 November 2003.

Abstract

In the last few years biometrics has emerged as an extremely useful technology for authenticating users. It has become more relevant given the security related issues presently plaguing our society. In this work we have concentrated on online-signatures and fingerprints, and how they can be used for user authentication.

  The work involved interfacing a digitizing tablet and a fingerprint sensor to the computer. The raw data obtained from the devices has to be processed and features extracted. Once the user has registered, he can approach the system for authentication at a later stage. We have also created a 30 user biometric database for use in any future research effort.

  In this work we have implemented a multi-expert on-line signature verification technique. For this 3 different algorithms have been implemented. For a user to be verified his signature has to be proven to be genuine by at least 2 of the algorithms (a majority classifier). Such a method has been adopted to lessen the inherent drawbacks that may exist in any particular algorithm and achieve better accuracies. Two of the algorithms extract local features while the third one is basically a set of selected global features. Local features have been extracted at critical points in case of algorithm 1 and at smoothed and resampled points of the entire signature in algorithm 2. In algorithm 3, twelve robust global features have been extracted.

Development of a fingerprint verification system has been attempted. Once we have the raw fingerprint image from the sensor various preprocessing steps such as finding the gradients in the image, segmenting it, estimating the orientation field and normalization to improve the ridge clarity have been implemented. The resulting image is then convolved with 2 orientation adaptive filters that help in ridge detection and give us a binary image. This image is then thinned and relevant features (ridge endings and bifurcations) are extracted. These features are then stored in the database for future retrieval and authentication.

  Hence we have proved that biometrics can reliably be used for automated user authentication by implementing a real time system based on handwritten signatures and fingerprints.