Publication: Finger vein presentation attack detection using deep learning and taguchi method
Loading...
Date
2023-08
Authors
Teh, Ven Ting
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Recent research has highlighted the susceptibility of finger vein biometrics to presentation attacks. Consequently, a finger vein presentation attack detection system has been developed to distinguish forged finger vein images from genuine ones. This project aims to create a reliable finger vein presentation attack detection system using deep learning and the Taguchi method. The project utilizes two databases, namely Idiap Full Database and SCUT-SFVD, which contain real and spoof finger vein images for training, validation, and testing. The images were pre-processed to enhance quality and resized to the recommended input dimensions of 224×224 pixels (height×width) for the EfficientNet B0 architecture. The Taguchi method was then employed to optimize the EfficientNet network through experimental designs. Subsequently, the model was trained and tested, and its performance was evaluated using metrics such as APCER, BPCER, and ACER. The proposed model's performance will be compared to that of the model without using the Taguchi method, the benchmark VGG-16 model, and the EfficientNet model with five additional fully connected layers at the end of the project. The Taguchi-based model had 4,070,165 total parameters and achieved the lowest inference time of 31.6125ms at Idiap Full Database, while also demonstrating a relatively low inference time of 36.8627ms at SCUT-SFVD, all while maintaining an accuracy of 100%. In the end of the research, the proposed Taguchi-based model is proven to reduce the total parameters and inference time while maintaining 100% accuracy.