Paper: "Optimization of the decision threshold for application in Biometric systems under extreme conditions of Facial image verification"

Abstract: Abstract: The accuracy of biometric facial recognition systems depends on a well-defined decision-making threshold that controls the rate of false positive verification results of unregistered users (FAR) and false negative results of verification of registered users (FRR). The decision threshold defines the balance between the security of the system and a better user experience. Greater security means a reduction in the ability to accept unauthorized users (low FAR), while a better user experience means a reduction in the number of unauthorized user rejection errors (low FRR). In this paper, three adapted methods for determining the decision threshold in biometric systems are presented, the methods are adapted for application in extreme conditions, such as facial recognition of users with masks. Method 1 uses the Gaussian optimization process. Method 2 is used to create a decision threshold based on the defined relationship between the FAR and FRR errors. Method 3 is based on the Grid Search algorithm. The results of the research show the potential of these methods with the possibility to use one of three adapted methods of calculating the decision threshold, depending on the implementation.

Keywords: Threshold decision-making, biometric facial recognition, Bayesian optimization, Grid Search.

Introduction: Biometric facial recognition is a biometric modality that can be used in the process of biometric verification, especially in the context of highly demanding secure systems such as electronic payment systems or a system for accessing a military-security complex. While facial recognition technology offers many benefits, challenges arise such as determining the correct decision threshold that affects system performance, Accuracy, False Acceptance Rate (FAR), and False Rejection Rate (FRR). In this paper, three adapted methods for the process of calculating the decision threshold for application in biometric systems are presented. A Sejong database of masked faces of 64 people was used to evaluate these methods [1]. The images are divided into an 80% training group and a 20% test group. The traditional approach to determining the decision threshold is the average value of the distance between the biometric template of images of the same user's face and the average distance for different users. The results presented in the paper demonstrate the limitations of the traditional calculation approach, especially when applied in complex situations such as masked faces. Existing methodologies, such as Gaussian Process Optimization [2], and Grid Search [3] were adapted to explore the threshold within a defined range, while performance was measured by the Accuracy metric. The development of innovative methods for calculating the decision threshold improves the performance of the biometric system. The paper is organized in such a way that the introductory chapter is followed by a chapter with an overview of previous scientific research. The third chapter describes the adapted methodologies. The results of the research are in the fourth chapter, while the fifth chapter is given a conclusion and the sixth chapter contains a list of literature.

Table 1 shows how Method 1 achieves the highest accuracy thanks to Bayesian optimization, but since a high FRR value is obtained, some solutions will require the use of the second or third threshold calculation method. Method 2 allows the calculation of the threshold with a balanced FAR and FRR error, while Method 3 offers flexible performance but with lower accuracy.

References:

[1] Usman Cheema, Seungbin Moon, Sejong face database: A multi-modal disguise face database, Computer Vision and Image Understanding, Volumes 208–209, 2021, ISSN 1077-3142, https://doi.org/10.1016/j.cviu.2021.103218.
[2] Bıyık E, Huynh N, Kochenderfer MJ, Sadigh D.: Active preference-based Gaussian process regression for reward learning and optimization. The International Journal of Robotics Research. 2024; 43(5):665-684. doi:10.1177/02783649231208729
[3] Aratrika Ray-Dowling, Daqing Hou, Stephanie Schuckers, Abbie Barbir: Evaluating multi-modal mobile behavioral biometrics using public datasets, Computers & Security, Volume 121, 2022, 102868, ISSN 0167-4048, https://doi.org/10.1016/j.cose.2022.102868.
[4] M. Nasr, A. Piórkowski, P. M. Mah and F. E. Abd El-Samie, Cancelable Biometrics Based on Quaternion Adaptive Filter Optimization for Security Applications, 2024 6th International Conference on Computing and Informatics (ICCI), New Cairo - Cairo, Egypt, 2024, pp. 490-498,
doi: 10.1109/ICCI61671.2024.10485055.
[5] Vasanthi, K. Seetharaman: Facial image recognition for biometric authentication systems using a combination of geometrical feature points and low-level visual features, Journal of King Saud University - Computer and Information Sciences, Volume 34, Issue 7, 2022, Pages 4109-4121, DOI:10.1016/j.jksuci.2020.11.028.
[6] P. Annamalai: Automatic Face Recognition Using Enhanced Firefly Optimization Algorithm and Deep Belief Network, International Journal of Intelligent Engineering and Systems, 2020, 13, Intelligent Networks and Systems Society, DOI: 10.22266/ijies2020.1031.03
[7] MS Sannidhan, Jason Elroy Martis, KN Pallavi, Vinayakumar Ravi, H. L. Gururaj, Tahani Jaser Alahmadi: Genetic algorithms and deep learning for unique facial landmark-based key generation, Computers and Electrical Engineering, Volume 118, 2024, DOI:10.1016/j.compeleceng.2024.109427.
[8] T. Edwards and M. S. Hossain: Effectiveness of Deep Learning on Serial Fusion Based Biometric Systems, IEEE Transactions on Artificial Intelligence, vol. 2, no.1, pp. 28-41, 2021, doi: 10.1109/TAI.2021.3064003.
[9] P. Wang, W. H. Lin, K. M. Chao and C. C. Lo: A Face-Recognition Approach Using Deep Reinforcement Learning Approach for User Authentication, IEEE 14th International Conference on e-Business Engineering (ICEBE), Shanghai, China, 2017, pp. 183-188, DOI:
10.1109/ICEBE.2017.36.
[10] Ríos-Sánchez, B., Silva D.C., Martín-Yuste, N., Sánchez-Ávila, C.: Deep learning for face recognition on mobile devices. IET Biom., 109-117. 2019. DOI:10.1049/iet-bmt.2019.0093
[11] Hadis Heidari, Abdolah Chalechale: Biometric authentication using a deep learning approach
based on different level fusion of finger knuckle print and fingernail, Expert Systems with Applications, Vol. 191, 2022, ISSN 0957-4174, https://doi.org/10.1016/j.eswa.2021.116278.

Reference:
Nenad BADOVINAC: "Optimization of the decision threshold for application in Biometric systems under extreme conditions of Facial image verification", INTERNATIONAL Conference "Annual conference on Challenges of Contemporary Higher Education" (1 ; Kopaonik ; 2025). Book of proceedings / 1. International Conference "Annual conference on Challenges of Contemporary Higher Education", ISBN-978-86-6211-150-0