Hybrid Biometric Person Authentication Using Face and Voice Features

In this paper, a hybrid person authentication prototype integrating multiple biometric devices is presented. This prototype is based on several levels of abstractions: data representations, vectors and classifiers. Frontal face and text -dependent voice biometrics are chosen to authenticate a user. For each of the biometric feature, an extractor, a classifier and a simple negotiation scheme have been designed. An extractor is made up of a sequence of operators which themselves are made up of signal processing and image processing algorithms. The face information is extracted using moments and the short speech information is extracted using wavelets. The extracted information, called vectors, is classified using two separate multi-layer perceptrons. The results are combined using a simple logical negotiation scheme. The prototype has been tested and evaluated on real-life databases.