Real-Time Client Authentication in Cyberspace Using the GLBPD and Three-Stream 3D-CNNs With the LSTM

• Published in: IEEE access 2024, Vol.12, p.151261-151274
• Main Authors: Feghhi, Mahmood Mohassel, Esmaeili, Vida
• Format: Article
• Language: English
• Subjects: Authentication; Biometrics; CASME and CASME 2 and MBGC datasets; Convolutional neural networks; Cyberspace; Face recognition; Feature extraction; GLBPD; Iris recognition; Long short term memory; Real-time systems; sequential iris images; three-stream 3D-CNNs with LSTM
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2024.3479718

Iris is the most accurate biometrics for authentication in cyberspace. Since it is unavailable for other persons, it creates more dependability to maintain national security. Also, it remains constant over time. In this paper, we propose combining the Gabor filter, the local binary pattern on diagonal planes, and three-stream 3D convolutional neural networks (3D-CNNs) with the long short-term memory (LSTM) for authentication from sequential iris images. We called it the GLBPD and three-stream 3D-CNNs with LSTM. First, the pre-processing techniques including the motion magnification, face alignment, eye detection, and cropping eye region are done. Second, the GLBPD extracts the iris features. Third, the sequences of the GLBPD maps are fed to three 3D-CNNs with the LSTM at the same time. Finally, the best result is reported by voting. This method is not only fully applicable in real-time, but also produces a high precision rate. Using the proposed method, the F1 score, precision, and accuracy are obtained 0.998, 1, and 0.999, respectively. Also, the Equal Error Rates (EER) of our method is very low.