A deep learning-based multimodal biometric system using score fusion

Abstract

Recent trends in artificial intelligence tools-based biometrics have overwhelming attention to security matters. The hybrid approaches are motivated by the fact that they combine mutual strengths and they overcome their limitations. Such approaches are being applied to the fields of biomedical engineering. A biometric system uses behavioural or physiological characteristics to identify an individual. The fusion of two or more of these biometric unique characteristics contributes to improving the security and overcomes the drawbacks of unimodal biometric-based security systems. This work proposes efficent multimodal biometric systems based on matching score concatenation fusion of face, left and right palm prints. Multimodal biometric identification systems using convolutional neural networks (CNN) and k-nearest neighbors (KNN) are proposed and trained to recognize and identify individuals using multi-modal biometrics scores. Some popular biometrics benchmarks such as FEI face dataset and IITD palm print database are used as raw data to train the biometric systems to design a strong and secure verification/identification system. Experiments are performed on noisy datasets to evaluate the performance of the proposed model in extreme scenarios. Computer simulation results show that the CNN and KNN multi-modal biometric system outperforms most of the most popular up to date biometric verification techniques.