Application of Depthwise Separable Convolutional Neural Network for Distorted Fingerprint Images

  •  Ramesh Chandra Sahoo et al.

Abstract

Biometric fingerprint authentication of an individual is a well-known pattern recognition problem and is widely used as authentication technologies such as accessing mobile devices, banking sector authentication, personal authentication and many more. The accuracy of classification and recognition is more important when these fingerprint images gets distorted or noisy due to various reason. In this study, a deep learning approach of convolutional network named Xception model is presented which involves depthwise separable convolutions for fingerprint classification and recognition with different noise percentage. The accuracy of classification and recognition rate is measured for 10 to 50 percentages of noisy fingerprint images on FVC2000, FVC2002 and FVC2004 fingerprint databases. These noises are explicitly introduced by our algorithm for different percentages and their recognition rate is measured with the original training data. The results of recognition obtained with this proposed model are of state-of-art and compared with different models discussed in the literature. In this implementation, we achieved 95.40% classification accuracy for Henry type classes and a accuracy of 97.5% was achieved for individual wise classification with Xception model

Published
2019-12-12
How to Cite
et al.. R. C. S. (2019). Application of Depthwise Separable Convolutional Neural Network for Distorted Fingerprint Images. International Journal of Control and Automation, 12(6), 448 - 455. Retrieved from https://sersc.org/journals/index.php/IJCA/article/view/2964