Longitudinal Study of Facial Biometrics at the BEZ: Temporal Variance Analysis

Mathias Schulz \sthanks [email protected], Pia Funk, Markus Rohde, Norbert Jung, Robert Lange, Alexander Spenke, Florian Blümel, Ralph Breithaupt, Gerd Nolden · 2025

Read Longitudinal Study of Facial Biometrics at the BEZ: Temporal Variance Analysis by Mathias Schulz \sthanks [email protected], Pia Funk, Markus Rohde, Norbert Jung, Robert Lange, Alexander Spenke, Florian Blümel, Ralph Breithaupt, Gerd Nolden (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-06858.txt.

Introduction

Mathias Schulz and co-authors report a two-and-a-half-year study of face recognition drift in a 2025 paper. Over four hundred volunteers of varied backgrounds visited a German test center on a set schedule. Their faces and fingers were captured with several devices under controlled conditions. The local database holds over two hundred thousand biometric samples, handled under European privacy law.

Modern face algorithms scored each visit against the volunteer's earlier records. Scores wobbled more from one day to the next than across the whole study span. Day-level noise, from light, mood, or skin state, may swamp slow aging effects. Test centers, the authors argue, must sample the same people across many days to grade systems fairly.

Catalog record

CREATORMathias Schulz \sthanks [email protected], Pia Funk, Markus Rohde, Norbert Jung, Robert Lange, Alexander Spenke, Florian Blümel, Ralph Breithaupt, Gerd Nolden
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE38 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2507-06858

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Longitudinal Study of Facial Biometrics at the BEZ: Temporal Variance Analysis Mathias Schulz \sthanks [email protected], Pia Funk, Markus Rohde, Norbert Jung, Robert Lange, Alexander Spenke, Florian Blümel, Ralph Breithaupt, Gerd Nolden Abstract This study presents findings from long-term biometric evaluations conducted at the Biometric Evaluation Center (BEZ) . Over the course of more than two and a half years, our ongoing research with over 400 participants representing diverse ethnicities, genders, and age groups were regularly assessed using a variety of biometric tools and techniques at the controlled testing facilities. Our findings are based on the General Data Protection Regulation (GDPR) - compliant local BEZ database with more than 238,000 biometric data sets categorized into multiple biometric modalities such as face and finger. We used state-of-the-art face recognition (FR) algorithms to analyze long-term comparison scores. Our results show that these scores fluctuate more significantly between individual days than over the entire measurement period. These findings highlight the importance of testing biometric characteristics of the same individuals over a longer period of time in a controlled measurement environment and lays the groundwork for future advancements in biometric data analysis. 1 Introduction Obtaining statistically significant results has long been a challenge in biometric system evaluations. Traditional approaches often involved conducting single-session tests with as many participants as possible. However, this method made it difficult to identify meaningful statistical influences, as individual variances and uncontrolled conditions compromised the reliability and generalization of the results. A long-term evaluation approach offers a solution to this problem by enabling systematic and controlled testing over extended periods. By maintaining controlled and adjustable conditions, several key advantages can be realized. First, the same systems can be tested with the same individuals over time, enabling robust longitudinal analysis, to evaluate the aging processes of people and systems. Second, different systems can be evaluated either simultaneously or within a short time frame under identical conditions, which facilitates direct comparisons. Third, newer versions of a system can be assessed in relation to their predecessors, providing insights into technological progress. In addition, extensions and modifications of existing systems, as well as the development of entirely new systems, can be evaluated in a structured manner. Finally, usability aspects can be examined thoroughly under these consistent conditions. Despite the evident benefits of long-term evaluations, current research lacks extensive datasets that reflect such conditions. Most studies rely on short-term assessments, failing to capture the effects of continuous system use, adaptation, and potential performance fluctuations over time. This study aims to bridge this gap by establishing a comprehensive long-term evaluation framework that systematically examines system performance, usability, and user adaptation over an extended period. 2 Related Work In the field of biometric evaluations, traditional methodologies have predominantly relied on single-session data collection, where subjects are enrolled and authenticated within a limited time-frame. This approach, while practical, often fails to account for temporal variations in biometric traits, potentially impacting the long-term reliability and accuracy of biometric systems. However, a few research institutions worldwide addressed this issue by investigating the long-term stability of biometric traits and evaluating biometric systems over extended periods in their studies. These studies provide valuable insights into the effects of aging and repeated exposures on biometric performance. The National Biometrics Laboratory at Israel’s National Cyber Directorate (INCD) has been instrumental in advancing biometric research, focusing on the long-term reliability of biometric systems. Their work contributes to national security measures and the development of robust biometric identification systems [ 1 ] . In the United States, the Biometrics Research Group at Michigan State University (MSU) [ 2 ] has conducted extensive longitudinal studies on both fingerprint and facial recognition systems: A significant study by L. Best-Rowden and A. K. Jain analyzed the permanence of facial features over time [ 3 ] . Utilizing large-scale mugshot databases, the researchers applied mixed-effects regression models to assess the degradation in genuine similarity scores as the time between enrolled and query images increased. Their findings indicated a measurable decline in recognition accuracy over extended periods, underscoring the necessity for longitudinal data to enhance the robustness of facial recognition systems. Yoon and Jain conducted an extensive longitudinal analysis of fingerprint recognition, examining records from 15,597 subjects over a span of up to 12 years [ 4 ] . The study revealed a significant decrease in genuine match scores correlating with longer time intervals between fingerprint captures. Despite this decline, the overall recognition accuracy remained stable, provided the fingerprint images were of high quality. Both of these studies highlight the importance of considering temporal factors in biometrics to maintain system reliability. However, the temporal distribution of data in these studies is often highly variable, as they rely on operational datasets such as mugshots, resulting in inconsistent intervals between recordings. Consequently, while such datasets offer valuable insights into long-term biometric performance, they may lack the controlled conditions and consistent measurement intervals required for specific precise temporal analysis. 3 Methodology 3.1 Participant acquisition This study was structured to allow repeated measurements of the same participants over an extended period to evaluate biometric characteristics under controlled long-term conditions. The measurements took place in the BEZ , which is located directly on the Bonn-Rhein-Sieg University (H-BRS) campus. Its central location proved to be highly advantageous for participant accessibility. Students and staff members were able to conveniently stop by during breaks, enabling frequent and flexible participation. Building on this logistical advantage, a variety of recruitment strategies were employed to ensure a diverse and representative sample. Flyers were distributed across key locations such as cafeterias and faculty buildings. Additionally, university-wide emails were sent to inform students and staff about the study and encourage participation. Posters highlighting the study’s objectives and participation requirements were strategically placed in high-traffic areas, including hallways and department bulletin boards. To further raise awareness, brief presentations were given in selected lectures, providing students with a direct opportunity to ask questions and to learn about the study. Moreover, a promotional advertisement was displayed on the large LED screen above the main entrance of the H-BRS , ensuring high visibility to both students and faculty members. To create an additional incentive for participation in the evaluations, vouchers are issued to the subjects after several visits. 3.2 BEZ evaluation process An additional important requirement, irrespective of the application, arises from the GDPR : the processing of personal data inherent in all biometric procedures requires special measures to ensure sufficient data protection. Figure 1: GDPR compliant BEZ evaluation process. 1. Enrollment: declaration of consent and biometric reference data is recorded. 2. Identification: autonomous check-in via biometric reference and participant to evaluation. The biometric database used in this study was constructed based on the collected data from recruited participants under controlled conditions. The GDPR compliant evaluation process is divided into two main phases, as illustrated in Fig. 1 : First, each subject underwent an initial enrollment phase to establish reference biometric data and to familiarize themselves with the study procedures. At the beginning of the enrollment session, participants attended a brief introductory session, which included a short guided tour of the testing environment and a detailed explanation of the study’s objectives, methodology and privacy policies. Then participants signed an informed consent form, ensuring voluntary participation and compliance with ethical

Read the complete text