Convolutional neural network-based feature extraction using multimodal for high security application

An efficient biometrics-based security system is the prime need in modern security industry. Biometric modalities are unique features of any human being based on which a computer system can recognise, authenticate or verify a person. In this paper we propose a convolutional neural network-based face, fingerprint, palm vein identification system. Main purpose of this paper is to propose a convolutional neural network with minimum layers for face, fingerprint and palm vein, achieving high accuracy and reducing the complexity. The network is of two convolutional layers, two ReLU layers and two Maxpooling layesr with ten hidden layers in Fully connected layer. The dataset of 4500 images is generated for all the modalities. Dataset images are used for 60% training, 10% validation and testing 30%. Proposed CNN architecture's accuracy is 95% for face, 94% for fingerprint and 99% palm-vein. The CNN used with minimum layers has performed consistently for all the biometric modalities maintaining good accuracy.