Advanced Multimodal Fusion for Biometric Recognition System based on Performance Comparison of SVM and ANN Techniques

In a multimodal biometric system, an efficient fusion method is necessary for combining information from various single modality systems. The score level fusion is used to combine several biometric features derived from different biometric modalities. Three biometric characteristics are considered in this study: Face, fingerprint and Voice. Classification methods represent also the basis of important recognition accuracy improvements. The artificial neural networks (ANN) and support vector machines (SVM) are considered as an excellent technique for classification. This paper presents a comparison of multimodal biometric recognition performances based on some methods that have been successfully applied using the fusion of scores. After exploring each modality (face, fingerprint and voice), we recovered three similarity scores. These scores are then introduced into two different classifiers: ANN and SVM. Experimental results demonstrate that a multimodal biometric system provides better performances than those using just one modalities system. Comparison of support vector machine and ANN based on score-level fusion methods is obtained and demonstrates that an average recognition rate(ARR=57.69%) is obtained using ANN. While fusion based on SVM gives an ARR= 63.31%.