Comparative Performance of Convolutional Neural Networks Architecture for Face Biometric Authentication System

• Published in: International Conference on Computing, Engineering, and Design (Online) pp. 1 - 4
• Main Authors: Mat Ariff, Noor Azwana, Ritahani Ismail, Amelia, Aziz, Normaziah Abdul
• Format: Conference Proceeding
• Language: English
• Published: IEEE 28.07.2022
• Subjects: Authentication; Biometric Authentication; Computer architecture; Convolutional Neural Networks; Deep learning; Error analysis; Face recognition; Image recognition; Passwords
• ISSN: 2767-7826
• DOI: 10.1109/ICCED56140.2022.10010512

Biometric authentication plays a vital role nowadays compared to password or token-based authentication. There are a lot of methods for biometric authentication algorithms that have been proposed but it can be said that the Deep Learning method give much more reliable and secure compared to other methods specifically Convolutional Neural Networks (CNN) for face recognition. Therefore, this paper will review the performance of top CNN architectures which are LeNet, AlexNet, VGGNet, GoogleNet, and ResNet by using the proposed face dataset of 7 celebrity classes where each class has 35 images that have been collected from Google Images. Data augmentation has been performed to increase the size of the dataset before it was fed into the CNN model. The experiment shows that AlexNet shows promising results compared to the other architectures on the proposed dataset.