Automated Authentication Using Hybrid Biometric System

A highly reliable biometrie authentication system can be realised by using multiple biometrie models. In this study, a framework that makes use of signal- and image-processing algorithms, together with pattern recognition techniques, is applied to solve the problem of biometrie pattern recognition in a unified way. In general, this problem can be broken down into the following taxonomy: sensors, extractors, experts and the supervisor. Using this general schema, biometrie systems with similar fundamental problem characteristics can be processed. According to the product law of reliability, a distributed (or parallel) system is more reliable than a linear system. Inspired by the idea of parallelism, ensemble methods and the notions of multi-sample and multi-model are studied. Based on the proposed framework, a hybrid biometrie authentication prototype that makes use of upright frontal face-scans and text-dependent voice-scans is implemented. This prototype has been tested on a real-life database in our laboratory with encouraging results. We show that multi-sample multi-model biometrie approach is more reliable than other existing combination models (single-sample single-model, single-sample multi-model or multi-sample single-model). From the application point of view, we have identified four categories of biometrie application according to several criteria: security (or accuracy) versus convenience (ease-of-use and non-intrusiveness), traffic flow and cost. We propose that the hybrid biometrie approach as an effective alternative when no other single-model biometrie approach can satisfy both the user (i.e. ease-of-use and non-intrusiveness) and technical (i.e. cost and accuracy) constraints at the same time.