Multimodal Person Authentication using Qualitative SVM with Fingerprint, Face and Teeth Modalities

Multimodal biometrics systems are becoming increasingly efficient over the unimodal system, especially for the securing handheld devices. However, the challenge with this authentication system is the relative degradation of the biometric modalities involved in the development and test data respectively. To overcome this problem, in this paper we propose a novel Qualitative Support Vector Machine (SVM) classifier with Face, teeth, and fingerprint as biometric traits. The test scores of individual modalities are adjusted according to their relative quality and then passed to binary SVM classifier. The experiments were conducted over a database collected from 20 individuals with three instances of all the three traits. The performance analysis of the fusion techniques revealed that the Equal Error Rates (EER) of 1.22%, 1.46%, and 1.88% for the qualitative SVM, raw score SVM and weighted summation rule classifiers respectively. On the other hand, the equal error rates for unimodal systems are 7.4%, 5.09% and 4.6% for teeth, face and fingerprint biometrics traits respectively. Hence, we confirmed that the proposed qualitative SVM method outperformed other raw score fusion techniques and unimodal classifiers.