Manuscript received October 28, 2022; revised December 14, 2022; accepted May 4, 2023.
Abstract—Security-based applications seek face biometrics as an integral part of the biometrics system which is susceptible to spoof attacks. A malicious person can gain unauthorized access to any system by displaying a picture or video of the registered user’s face. Anti-spoofing techniques are becoming increasingly crucial in the face biometric authentication systems. Convolution Neural Networks (CNN) have recently gained popularity in important computer vision application areas, encouraging their usage for face spoof detection. Even though deep networks are more resistant, such designs require expensive computational training. Also, the adoption of deep CNN architectures for face anti-spoofing applications has been constrained by the lack of sufficient training data that the existing spoof datasets can offer. Also trained models to lack generalizability concerning unknown data domains, and are not robust enough to handle unseen attacks. We propose a Transfer Learnt Anti-spoof (TLA) framework in this paper, to induce and improve generalizability and accuracy in spoof detection. TLA framework consists of two Convolution neural networks namely ResNet-34 and MobileNetV2. Here we pre-train these two CNN models on a larger dataset at the base. Then a dense classification layer is formed to classify the features obtained from the previous convolutional base, into the real and spoofed faces. The TLA Framework was applied efficiently over the NUAA Photo Imposter dataset and the models within the framework exhibited the highest accuracy of 99.76% and 99.60% respectively for spoof detection and tests demonstrate that TLA outperforms the state-of-the-art techniques.
Index Terms—Transfer learnt anti-spoof framework, transfer learning, ResNet-34, MobileNetV2, face anti-spoofing
H. Vinutha is with Department of Computer Science and Engineering, BMS Institute of Technology and Management, Bengaluru, India and Department of Computer Science and Engineering, Dr. Ambedkar Institute of Technology, Bengaluru, India.
G. Thippeswamy is with Department of Computer Science and Engineering, B M S Institute of Technology and Management, Bengaluru, India.
*Correspondence: vinuthah@gmail.com (H.V.)
[PDF]
Cite:H. Vinutha and G. Thippeswamy, "TLA Framework—A Transfer Learning Based Approach for Face Anti-spoofing," International Journal of Computer Theory and Engineering vol. 15, no. 4, pp. 143-151, 2023.
Copyright © 2023 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).
