A Framework for Biometric Authentication based on Decision Level Fusion

Unimodal systems are still facing challenges in authentication though there are considerable advances in recent years. Some of the challenges can be handled by designing a multimodal biometric system. A decision fusion framework for selected biometrics has been proposed and developed. The basic idea here is to fuse the decisions obtained from the individual matchers for face, iris, and fingerprint and signature. Each biometric decision was evaluated using hamming classifiers. The individual decisions from the all modalities were further combined with straightforward the AND logic rule to obtain the final decision. Proposed methodology employs AND logic for a satisfactory level of security. Person is authenticated as a genuine if and only if all biometrics modalities result into positive authentication. An evaluation of decision fusion method based on AND rule-based approach has been presented in this work. To evaluate the performance of the proposed system, we have performed combination of Casia database, FVC2004 database with signature databases as UCOER, Caltech database, and face databases as ORL, Yale, IIT Female database. The experimental results indicate that the decision-level fusion outperforms unimodal biometrics system in terms of different error rates and GAR. We have reported better results as FAR = 0% with FRR = 0.0110% with GAR = 99.89%. Experimental results prove that proposed fusion algorithm excels in performance than other decision approaches in literature.