CoChat
Sign Up
Cited by 5Face recognition and analysisBiometric Identification and SecurityFace and Expression Recognition

The DOI matches its Crossref record; 20 of the 20 references it cites were found in a registry or an index; 3 of the 8 sources its claims cite were checked for support.

Longitudinal Evaluation of Child Face Recognition and the Impact of Underlying Age

Open access
Singh, Surendra, Bahmani, Keivan, Schuckers, Stephanie·2024·2024 IEEE International Joint Conference on Biometrics (IJCB)
DOI 10.1109/ijcb62174.2024.10744505
Open this paper in CoChat — free accountOpen PDF

Abstract

The need for reliable identification of children in various emerging applications has sparked interest in leveraging child face recognition technology. This study introduces a longitudinal approach to enrollment and verification accuracy for child face recognition, focusing on the YFA (Young Face Aging) database collected by Clarkson University’s CITeR research group over an 8-year period, at 6-month intervals. The dataset includes children ranging from 3 to 18 years of age, comprising 330 subjects with an average of 6 data collections per subject. Our research aims to comprehensively evaluate the performance of state-of-the- art face-matching techniques on the YFA database, assessing the feasibility of recognizing children's faces upon initial enrollment and verifying their identity longitudinally at 6-month intervals. We conduct a comprehensive analysis of the system’s accuracy considering multiple age groups. We also investigate the temporal degradation of face recognition accuracy over time. Notably, when comparing the initial enrollment image with longitudinal images over an 8-year period, we observe a decrease in accuracy. The average TAR across all age groups is 98.52% with a FAR of 0.1% with a 2-year age verification gap and drops to 95.68 with a 4-year age gap. However, this rate decreases to 87.24% after a time difference of 6 years and further drops to 71.32% with a time difference of 8 years. The highest drop in accuracy was noticed in the age group of (3-5) years old children and the lowest in (5.5-7) years old. By addressing the challenges and opportunities in child face recognition, this research contributes significantly to the advancement of technology for identifying missing or abducted children and other critical applications requiring dependable biometric recognition in children.

Overview

AI snapshot

Written from the paper’s full text.

TL;DR

This study longitudinally evaluates child face recognition on the YFA database using MagFace, tracking verification accuracy as children age from 3 to 18 over up to 8 years. Accuracy declines with longer enrollment-verification gaps and varies by age group, with the largest drop for ages 3–5 and best performance for ages 5.5–7.

Key findings

  • With MagFace at 0.1% FAR, average TAR across all age groups was 98.52% at a 2-year age verification gap, 95.68% at 4 years, 87.24% at 6 years, and 71.32% at 8 years.
  • The largest accuracy decline occurred in the 3–5 age group, where TAR dropped to 63.1% for a verification age gap of 6.5–8 years; the 5.5–7 group had the highest TAR at 80.2%, while the 7.5–9 group dropped to 65.6%.
  • Table 4 reports MagFace at 95.48% TAR at 0.1% FAR for the extended YFA with an 8-year longest time gap and 6-month interval; prior YFA with a 3-year gap had MagFace at 94.9%.
  • Accuracy did not show consistent trends with increasing age within enrollment age groups.

Methods

  • Dataset: YFA, collected by Clarkson University’s CITeR research group over 8 years at 6-month intervals; 330 subjects aged 3–18, 3,831 samples, average 6 collections per subject.
  • Protocol: longitudinal enrollment and verification; any image can be enrollment, subsequent collections are verification, with increasing time intervals; subjects categorized into age brackets.
  • Matcher/analysis: MagFace face-matching algorithm; performance measured as TAR at 0.1% FAR across age groups and time gaps.

Limitations

  • Stated: uneven subject distribution across ages 3–18; variations in facial movement, angle, and expression; lighting inconsistencies across classroom collections; appearance differences such as glasses and caps.
  • Stated: lack of demographic diversity, particularly ethnicity, limits generalizability; only MagFace was used, and other matchers or fine-tuning could change results.
  • Stated: findings are from a controlled environment and may be lower in real-world scenarios due to image quality, environmental conditions, and operational constraints. Evident from text: the methodology section is fragmentary in the provided text, but the paper’s own limitations are otherwise explicit.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1109/ijcb62174.2024.10744505

Cites

20 of 20 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

3 of 8 cited sources checked · 2 supported · 0 flagged · accuracy not yet measured

  • Premisecitation

    Yau et al. demonstrated significant variations in authentication robustness between age groups.

    “A study conducted by Yau et al. [18], significant variations in authentication robustness between age groups are demonstrated.”
    Quote in the paperNot a substantial claim1 of 1 sources checked
  • Premisecitation

    Bahmani et al. found a significant and consistent decline in match scores with increasing age gaps between gallery and probe images in children, even over short intervals such as 6 months.

    “The research findings reveal a significant and consistent decline in match scores, with increasing age gaps between gallery and probe images in children, even over short intervals such as 6 months.”
    Quote in the paperSubstantial claim1 of 1 sources checked
  • Premisecitation

    Bahmani et al. achieved 98.3% and 94.9% TAR at 0.1% FAR over 6 and 36 months age gaps with MagFace.

    “With MagFace they achieved 98.3% and 94.9% TAR at 0.1% FAR over 6 and 36 months age gaps.”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [2] source

      Table III (C. Face Detection and Recognition Models) reports MagFace TAR at 0.1% FAR as 98.2% at 6 months, 98.3% at 12 months, and 94.9% at 36 months. The 36-month figure matches the claim, but the claimed 6-month figure belongs to the 12-month age gap. (LLM, conf=0.90)

      Experimental — accuracy not yet measured
  • Findingown data

    The average TAR decreases to 87.24% after a time difference of 6 years.

    “However, this rate decreases to 87.24% after a time difference of 6 years”
    Quote in the paperSubstantial claim
  • Findingown data

    The lowest drop in accuracy was in the age group of (5.5-7) years old.

    “and the lowest in (5.5-7) years old.”
    Quote in the paperNot a substantial claim
  • It may state two separate claims as one. 7
  • Probably not a substantial claim. 4

  • It may state two separate claims as one.

    Findingnone given

    The YFA dataset covers ages 3-18 and increases in time between enrollment and verification for 6-month increments up to 8 years.

    “YFA covers ages 3-18, as well as increases in time between enrollment and verification for 6month increments up to 8 years.”
    Quote in the paperSubstantial claim

  • It may state two separate claims as one.

    Findingown data

    Previous research utilized the YFA database with collection periods of up to 3 years, whereas this study expanded the database to include collection periods of up to 8 years.

    “Previous research utilized the YFA database with collection periods of up to 3 years, whereas we expanded our database to include collection periods of up to 8 years.”
    Quote in the paperSubstantial claim

  • It may state two separate claims as one.

    Findingown data

    The YFA dataset includes 330 subjects with an average of 6 data collections per subject.

    “The dataset includes children ranging from 3 to 18 years of age, comprising 330 subjects with an average of 6 data collections per subject.”
    Quote in the paperSubstantial claim

  • It may state two separate claims as one.

    Findingown data

    The average TAR across all age groups is 98.52% with a FAR of 0.1% with a 2-year age verification gap.

    “The average TAR across all age groups is 98.52% with a FAR of 0.1% with a 2-year age verification gap”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim. The paper’s conclusions may not depend on it.

    Study design and conductnone given

    There are inconsistencies in lighting across collections.

    “However, there are still inconsistencies in lighting across collections.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim. The paper’s conclusions may not depend on it.

    Study design and conductnone given

    Disparities in subjects' appearance, such as some wearing glasses and caps, complicate the recognition process.

    “Moreover, disparities in subjects' appearance, such as some wearing glasses and caps, further complicate the recognition process.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim. The paper’s conclusions may not depend on it. It may state two separate claims as one.

    Premisecitation

    Siddiqui et al.'s CNN architecture achieves a rank-1 identification accuracy of 62.7% for single gallery newborn face recognition and 85.1% for single gallery toddler FR.

    “The CNN architecture proposed in the study achieves a rank-1 identification accuracy of 62.7% for single gallery newborn face recognition and 85.1% for single gallery toddler FR.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [15] source

  • It may state two separate claims as one.

    Premisecitation

    Deb et al. found that the accuracy of face recognition decreases over time, with 83.77% TAR at 0.1% FAR with a 1-year time-lapse, decreasing to 59.80% TAR after 3 years.

    “In this study, the researchers found out that the accuracy of face recognition decreases over time.Initially, they achieved an accuracy of 83.77% True Acceptance Rate (TAR) and 0.1% False Acceptance Rate (FAR) with a 1-year time-lapse, but this decreased to 59.80% TAR after 3 years.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [6] source

  • It may state two separate claims as one.

    Study design and conductnone given

    The distribution of subjects within the 3 to 18 years age range is uneven, potentially impacting the generalizability of findings.

    “Although the study encompasses children aged 3 to 18 years, the distribution of subjects within this age range is uneven, potentially impacting the generalizability of findings.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim. The paper’s conclusions may not depend on it.

    Study design and conductnone given

    Variations in facial movement, angle, and facial expressions were present despite attempts to control quality factors.

    “While the collectors attempted to control various quality factors, variations in facial movement, angle, and facial expressions were present.”
    Quote in the paperNot a substantial claim
10 of 10

  • The paper relies on this without support.

    There is a growing demand for reliable identification of children across various applications.

    “In recent years, there has been a growing demand for reliable identification of children across various applications, including missing children, border security, humanitarian, and health care.”
1 of 1

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • NITL's longest gap exceeds its collection period

    Minor

    Table 1 gives NITL a collection period of 1 year, but Table 4 reports a longest time gap of 2 years for the same dataset.

1 of 1

Citations

Cites

Flagged first, then by citations

Research works printed but not linked: 2

  1. ISO iso/iec tr 19795-3:2007), information technology biometric performance testing and reporting
  2. Huang, 2014, Labeled faces in the wild: Updates and new reporting procedures, 14

Cited by

Sorted by citations

Related

Metadata

Authors
Singh, S., Bahmani, K., & Schuckers, S.
Affiliations
Clarkson University (US)
ORCID
Venue
2024 IEEE International Joint Conference on Biometrics (IJCB)
Publisher
IEEE
Published
2024-09-15
Pages
1-9
DOI
10.1109/ijcb62174.2024.10744505
OpenAlex
W4404238897
Semantic Scholar
d49cc7f2f641395a0b4ce1f33179d3745b7afc35
Cited by
5
References
18
Citations by year
2026: 2 · 2025: 3
FWCI
0.72
Type
conference paper
Language
en
Open access
Licence
Locations
  • IEEE International Joint Conference on Biometrics (IJCB) (publishedVersion)
  • arXiv (Cornell University) (submittedVersion)
  • arXiv (Cornell University)
Topics
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Face recognition and analysis
  • Physical Sciences › Computer Science › Signal Processing › Biometric Identification and Security
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Face and Expression Recognition
Keywords
facial aging, face verification, face matching, age groups, biometric identification

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

A language model drew the claims from the paper on 2026-11-09, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-11-09 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review