Finger Vein Based Authentication using Deep Learning Techniques
Madhusudhan M V1, Udayarani V2, Chetana Hegde3
1Madhusudhan M V*, Computer Science Engineering, Presidency University, Bengaluru, India.
2Udayarani V, School of Computing and IT, Reva University, Bengaluru, India.
3Chetana Hegde, Senior Faculty of Data Science, Manipal ProLearn, Bangalore, India. 
Manuscript received on January 02, 2020. | Revised Manuscript received on January 15, 2020. | Manuscript published on January 30, 2020. | PP: 5438-5443 | Volume-8 Issue-5, January 2020. | Retrieval Number: E6890018520/2020©BEIESP | DOI: 10.35940/ijrte.E6890.018520

Open Access | Ethics and Policies | Cite | Mendeley
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC BY-NC-ND license (

Abstract: Security is one of the major concerns of current times. Biometric based methods are found to be more reliable and accurate in authenticating an individual. Hand-based biometric traits are proved to be easily accessible during data collection. Collecting, storing and processing biometric trait images of all the employees is always a challenge for larger organizations. Deep learning techniques come to rescue from such situations. In this paper, we propose a novel approach for authentication using finger-vein images. We use basic convolutional neural network (CNN) with transfer learning. The model has been pre-trained on various types of images available on ImageNet database through ResNet – 50 architecture. This pre-trained model has been then run through CNN model with appropriate number of hidden layers and activation functions. The optimizers and loss functions are used to achieve appropriate classification among the images. The simulation results of proposed model has shown 99.06% of accuracy in classifying an individual.
Keywords: Adam Optimizer, Categorical Cross Entropy Loss Function, Convolutional Neural Network, Deep Learning, Dropout, Early Stopping, Relu Activation Function, ResNet–50, Softmax Activation Function.
Scope of the Article: Deep Learning