Multimodal biometrics – Representation, optimization and kernelization

Abstract

Traditional biometric systems use a single source of information such as fingerprint, retinal characteristic, and signature for tracking an individual. With today’s technology, spoofing and duplication of such biometrics has become easier. Sometimes, the data collected may not be accurate due to cases like, use of contact lens by an individual, or water on the finger of the subject during the collection of biometric samples. Hence there is a need of advanced systems to tackle these problems.

The current project presents a method, which uses multiple biometrics of an individual, making recognition more robust as it is difficult to duplicate all the biometric samples of an individual accurately. The proposed scheme employs feature-level fusion for the multimodal biometric recognition. The scheme is made robust through optimization, then the quality measure of the data is analyzed as a part of the joint sparse recognition and finally the data is kernelized to reduce nonlinearity.