A Multimodal Biometric System for Secure User Identification Based on Deep Learning

Abstract

A multimodal biometric system utilizes more than one biometric modality of a person to relieve some of the shortcomings of a unimodal biometric system and improves its security. In this paper, we propose a novel deep learning approach for fusing the features extracted from the individual’s face and iris (left and right) to get a more secure biometric verification system. Firstly, we extract the facial and iris features separately using various convolutional neural network (CNN) models. Further, the feature vectors of the final CNN layers of both models are fused to achieve classification of individuals with improved performance. The proposed system is tested on the CASIA-Face V5 dataset for faces and IITD iris dataset for left and right irises. The results achieved prove the superiority of the proposed multimodal system. It is efficient, reliable, and robust as compared to unimodal biometric systems.