Aditya Chiduruppa¹ and Praveen Kumar Pandian Shanmuganathan²*

¹High School Junior, Lexington High School, Boston, MA, United States of America
²Computer Vision Scientist, Machine Learning Researcher, Florida Institute of Technology, Pittsburgh, PA, United States of America

*Corresponding Author: Praveen Kumar Pandian Shanmuganathan, Computer Vision Scientist, Machine Learning Researcher, Florida Institute of Technology, Pittsburgh, PA, United States of America.

Received: September 22, 2025; Published: September 29, 2025

Abstract

Facial recognition technology is increasingly utilized for user identification and authentication in various applications. However, several variations across the faces are presented in these systems. This made me wonder how important and prevalent biases are within these systems, namely across races, genders, and beyond. This study initially investigated biases in facial recognition algorithms, focusing on variations in accuracy based on ethnicity, skin tone, and gender. The research comprises two experimental phases. In the first phase, a dataset of facial images from 70,000 individuals, categorized into 11 distinct skin tone groups, was analyzed to evaluate recognition accuracy. Results revealed significant disparities across skin tones, confirming the presence of inherent algorithmic bias. In the second phase, an alternative analysis using the Illinois DOC dataset compared recognition accuracy and False Acceptance Rates (FAR) between African American and Caucasian groups under varying thresholds. To address these biases, Perspectum, a novel metric, will be developed to quantify errors and biases within facial recognition systems in a way that is practical for an enduser making important decisions in a border patrol or law enforcement scenario. Perspectum will be a metric that provides a number between 0 and 100 which can be seamlessly integrated into existing face recognition systems, offering a practical solution to mitigate bias.

Keywords: False Acceptance Rates (FAR); United Healthcare; Facial Recognition Technology (FRT)

References

1. “Racial Profiling”.
2. Kimery Anthony. “Limitations of FRT apparent in search for United Healthcare CEO’s killer”. BiometricUpdate.com (2024).
3. K Gates. “Can computers be racist?”. Juniata Voices 15 (2015).
4. GM Haddad. “Confronting the biased algorithm: the danger of admitting facial recognition technology results in the courtroom”. Vanderbilt Journal of Entertainment and Technology Law 23 (2020): 891.
5. K Krishnapriya., et al. “Issues related to face recognition accuracy varying based on race and skin tone”. IEEE Trans. Technol. Soc. 1.1 (2020): 8-20.
6. M Jmal., et al. “Classification of human skin color and its application to face recognition”. presented at the MMEDIA 2014: the Sixth International Conference on Advances in Multimedia, IARIA (2014).
7. Y Ban., et al. “Face detection based on skin color likelihood”. Pattern Recognition4 (2014): 1573-1585.
8. B Dhivakar., et al. “Face detection and recognition using skin color”. Presented at the 2015 3rd International Conference on Signal Processing, Communication and Networking (ICSCN), IEEE (2015): 1-7.
9. S Wehrli., et al. “Bias, awareness, and ignorance in deep-learning-based face recognition”. AI Ethics3 (2022): 509-522.
10. V Muthukumar., et al. “Understanding unequal gender classification accuracy from face images”. ArXiv Prepr. ArXiv181200099 (2018).
11. V Muthukumar. “Color-theoretic experiments to understand unequal gender classification accuracy from face images”. presented at the Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (2019).
12. P J Phillips., et al. “An other-race effect for face recognition algorithms”. ACM Trans. Appl. Percept. TAP 8.2 (2011): 1-11.
13. M Atay., et al. “Evaluation of gender bias in facial recognition with traditional machine learning algorithms”. presented at the 2021 IEEE Symposium Series on Computational Intelligence (SSCI), IEEE (2021): 1-7.
14. K Vangara., et al. “Characterizing the variability in face recognition accuracy relative to race”. presented at the Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (2019).
15. J Buolamwini and ID Raji. “Actionable auditing: investigating the impact of publicly naming biased performance results of commercial AI products”. Conference on Artificial Intelligence, Ethics, and Society, (2019).
16. P Grother., et al. “Face recognition vendor test (fvrt): Part 3, demographic effects”. National Institute of Standards and Technology Gaithersburg, MD (2019).
17. T Gwyn and K Roy. “Examining gender bias of convolutional neural networks via facial recognition”. Future Internet12 (2022): 375.
18. P J Phillips., et al. “Face recognition accuracy of forensic examiners, superrecognizers, and face recognition algorithms”. Proceedings of the National Academy of Sciences of the United States of America 24 (2018): 6171-6176.
19. “Flickr-Faces-HQ Dataset (FFHQ)”. NVlabs, (2019).
20. W Coleman., et al. “Updating the Fitzpatrick classification: the skin color and ethnicity scale”. Dermatology Surgery8 (2023): 725-731.
21. B Amos., et al. “Openface: A general-purpose face recognition library with mobile applications”. CMU School of Computer Science 2 (2016): 20.
22. A Geitgey. “face_recognition”. Adam Geitgey (2018).

Citation

Citation: Aditya Chiduruppa and Praveen Kumar Pandian Shanmuganathan. “Analyzing Bias in Face Recognition and Addressing it with the Perspectum Tool". Acta Scientific Computer Sciences 7.7 (2025): 20-30.

Copyright

Copyright: © 2025 Aditya Chiduruppa and Praveen Kumar Pandian Shanmuganathan. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.