Multimodal Authentication using Qualitative Support Vector Machines

This paper proposes an approach to enhancing the accuracy of multimodal biometrics in uncontrolled environments. Variation in operating conditions results in mismatch between the training and test material, and thereby affects the biometric authentication performance regardless of this being unimodal or multimodal. ne paper proposes a technique to reduce the effects of such variations in multimodal fusion. The proposed technique is based on estimating the quality aspect of the test scores and then passing these aspects into the Support Vector Machine either as features or weights. Since the fusion process is based on the learning classifier of Support Vector Machine, the technique is termed Support Vector Machine with Quality Measurement (SVM-QM). The experimental investigation is conducted using face and speech modalities. The results clearly show the benefits gained from learning the quality aspects of the biometric data used for authentication.