Year: 2025
Type: article
Abstract: Biometric technologies are fast becoming a requirement in security systems today, providing solutions where traditional means alone would not be adequate. This paper proposes FootprintNet, a Siamese network that utilizes pre-trained convolutional neural networks, specifically EfficientNet, MobileNet, and ShuffleNet, to improve the robustness and accuracy of footprint recognition. By learning the ability to identify fine distinctions between images of footprints, FootprintNet offers great biometric identification potential. Detailed analysis of the Biometric 220 × 6 Human Footprint dataset shows a true positive rate over 99% under various thresholds and a precision rate of 100% during training. Most importantly, this system is also applicable to newborn and infant identification, making it especially significant in medical settings, including hospitals and birthing clinics. Furthermore, the model sizes range from 7.8 MB (ShuffleNet) to 24.5 MB (EfficientNet), which makes FootprintNet deployable on low-computational-power devices—a highly desirable trait for mobile or high-security applications.
Source: The Journal of Supercomputing
Authors: Nadir İbrahimoğlu, Amjad Osmani, Ali Ghaffari, Faruk Baturalp Günay, Tuğrul Çavdar, Furkan Yıldız
Language: English
Subfield: Signal Processing
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $3,290
OpenAlex ID: W4409285675