CNN Hyperparameter Tuning Applied to Iris Liveness Detection
Gabriela Y. Kimura, Diego Rafael Lucio, Alceu S. Britto, David Menotti · 2020 · 32 citationsRead the paper
The iris pattern has significantly improved the biometric recognition field due to its high level of stability and uniqueness. Such physical feature has played an important role in security and other related areas. However, presentation attacks, also known as spoofing techniques, can be used to bypass the biometric system with artifacts such as printed images, artificial eyes, and textured contact lenses. To improve the security of these systems, many liveness detection methods have been proposed, and the first Internacional Iris Liveness Detection competition was launched in 2013 to evaluate their effectiveness. In this paper, we propose a hyperparameter tuning of the CASIA algorithm, submitted by the Chinese Academy of Sciences to the third competition of Iris Liveness Detection, in 2017. The modifications proposed promoted an overall improvement, with an 8.48% Attack Presentation Classification Error Rate (APCER) and 0.18% Bonafide Presentation Classification Error Rate (BPCER) for the evaluation of the combined datasets. Other threshold values were evaluated in an attempt to reduce the trade-off between the APCER and the BPCER on the evaluated datasets and worked out successfully.
What the full text shows
Shown in the lab- Combined dataset performance improvement over baseline CASIA algorithm. — 8.48% APCER and 0.18% BPCER (Combined datasets, threshold = 50)
- Lowest error rate achieved on the Clarkson dataset. — 4.18% APCER and 0% BPCER (Clarkson dataset, threshold = 50)
- Optimal learning rate for training. — 1e-5 (SpoofNet training)
- Optimal regularization parameters. — Weight decay 10-4, dropout 0.2 (SpoofNet training)
- Setup
- Four datasets (Clarkson, Warsaw, IIITD-WVU, Notre Dame) containing grayscale images (640x480 pixels) of live irises and spoof attacks (printed images, textured contact lenses).
- How
- The authors performed hyperparameter tuning on the CASIA SpoofNet algorithm, optimizing batch size, epochs, learning rate, and weight decay. They also evaluated various classification thresholds (30 to 90) to manage the trade-off between APCER and BPCER.
- Limitations
- Higher APCER on the Notre Dame dataset compared to the original CASIA algorithm. · Difficulty in classifying unknown spoofing attacks in testing subsets. · Trade-off between APCER and BPCER varies by dataset (e.g., Warsaw dataset responded differently to threshold changes than others).
Summarized by Gemma 4 from the open-access full text · 18 Oct 2026. Read it