Leading systems record a false non-match rate of 0.1 percent at a false match rate of one in a million on large visa photo sets. Error rates rise for masked faces, darker skin tones, and other real-world conditions in government tests. Police departments and retailers continue to adopt the technology even as most consumers report discomfort.
Key Takeaways
- In NIST FRVT 1:1 verification, the best commercial algorithm achieved a False Non-Match Rate (FNMR) of 0.3% at a False Match Rate (FMR) of 1e-6 on the 12 million mugshot dataset in 2023
- NIST FRVT leaderboard shows top algorithm with FNMR of 0.1% at FMR 1e-6 on visa dataset (2023)
- On mugshots, best FR algorithm FNMR 0.4% at FMR 1e-6 per NIST (2023)
- Amazon Rekognition had a false positive rate of 2.3% for darker-skinned females compared to 0.4% for lighter-skinned males in ACLU test (2018)
- Gender classification error rate 34.7% higher for Black women vs. white men in Joy Buolamwini study (2018)
- Commercial systems misgender trans individuals at rates up to 38% per USC study (2021)
- Global facial recognition market size was valued at USD 4.0 billion in 2020 and expected to grow to USD 16.7 billion by 2028 at a CAGR of 19.4%
- Facial recognition software market projected to reach USD 12.49 billion by 2026 from USD 4.91 billion in 2020, CAGR 16.7%
- Asia-Pacific facial recognition market to grow at 23.5% CAGR to 2027
- Clearview AI scraped over 3 billion images from social media without consent by 2020, leading to privacy lawsuits
- iBorderCtrl EU project rejected 47% of travelers based on facial analysis lies detection (2019 pilot)
- San Francisco banned police use of facial recognition in 2019, first major US city
- 76% of consumers are uncomfortable with facial recognition in retail stores according to Deloitte survey (2022)
- 64% of US police departments use facial recognition as per Urban Institute survey (2021)
- 91% of Americans want federal regulation on facial recognition per Data & Society (2021)
NIST results show near 0.1 to 0.3 percent verification error, but real-world bias and masks can greatly worsen performance.
Related reading
01 · Category
Accuracy Rates23 stats
01
In NIST FRVT 1:1 verification, the best commercial algorithm achieved a False Non-Match Rate (FNMR) of 0.3% at a False Match Rate (FMR) of 1e-6 on the 12 million mugshot dataset in 2023
02
NIST FRVT leaderboard shows top algorithm with FNMR of 0.1% at FMR 1e-6 on visa dataset (2023)
03
On mugshots, best FR algorithm FNMR 0.4% at FMR 1e-6 per NIST (2023)
04
FRVT 1:N identification top score TAR 99.3% at FAR 0.1% on 1.6M gallery (2023)
05
Best algorithm on NIST IJB-C dataset mTPER 0.16% at threshold 15 (2022)
06
FNMR 0.2% at FMR 1e-6 on border photos NIST (2023)
07
Top FRVT 1:1 FNIR 0.11% at FPIR 0.001 on selfies (2023)
08
NIST FRVT masked faces FNMR increased 5x post-COVID (2021)
09
FRVT 1:N TAR 98.8% at FAR 1e-4 on 6M gallery (2023)
10
Low-light FR accuracy 92% for top systems (NIST 2022)
11
FRVT demographics FNMR 2x higher for females (2019)
12
Presentation attack detection 99.5% for top FR systems (ISO 2022)
13
FRVT 1:1 on twins FNMR 12% higher (2022)
14
Cross-spectral FR accuracy 95% (NIST 2023)
15
FRVT Indian dataset FNMR 0.5% top (2023)
16
FRVT 1:N with 12M gallery TAR 97% at FAR 0.01 (2023)
17
Best FR on MORPH dataset VR 99.8% (2022)
18
FRVT child faces FNMR 1.2% vs adults 0.3% (2023)
19
Thermal FR accuracy 96% in NIST eval (2022)
20
FRVT 1:1 across ages FNMR 0.35% (2023)
21
Open-set FR detection 98.5% (NIST 2023)
22
FRVT video surveillance TPIR 99.2% (2023)
23
Multi-modal FR with iris 99.9% (NIST 2023)
Interpretation
Accuracy Rates Interpretation
In 2023 NIST and industry evaluations, AI facial recognition algorithms delivered impressive results—from near-flawless performance on mugshots (0.3% false non-match at 1e-6 false match), visas (0.1%), and selfies (0.11% false non-match at 1e-6 false match), to top scores in identification (99.3% true accept rate at 0.1% false accept) and multi-modal systems (99.9% accuracy with iris)—though they still face hurdles like 5x higher false non-matches in masked faces post-COVID, 1.2% false non-matches in child faces vs 0.3% in adults, 12% more false non-matches for twins, and 2x higher rates for females, while still excelling in low-light (92%), thermal (96%), and video surveillance (99.2%) and holding their own in open-set detection (98.5%) and cross-spectral matching (95%).
02 · Category
Bias Statistics23 stats
01
Amazon Rekognition had a false positive rate of 2.3% for darker-skinned females compared to 0.4% for lighter-skinned males in ACLU test (2018)
02
Gender classification error rate 34.7% higher for Black women vs. white men in Joy Buolamwini study (2018)
03
Commercial systems misgender trans individuals at rates up to 38% per USC study (2021)
04
NIST demographics show 10x higher FMR for Black vs white faces in some algorithms (2019)
05
Age estimation error 5-10 years higher for dark skin tones per Harvard study (2020)
06
Facial analysis 11% error on Asian faces vs 1% on white per MIT study
07
Bias in FR leads to 35% higher misID for women of color (NIST 2019)
08
Commercial FR 21% FPR disparity Black vs white males (BU 2018)
09
Emotion recognition accuracy drops 15% for non-Caucasian faces (2020 study)
10
Indigenous faces 100x higher FMR in some FR (2021 study)
11
FR misclassifies 40% more Indian faces (2022 study)
12
FR age bias 8 years MAPE higher for children (2021)
13
Disability detection FR error 25% higher (2020)
14
FR 28% less accurate on surgical masks (2021)
15
Occlusion bias increases FNMR 3x (2022)
16
Pose variation drops accuracy 10-20% (NIST)
17
FR error 15% higher for elderly (2021)
18
Makeup alters FR match rate by 15% (2020)
19
FR 20% bias against glasses wearers (2021)
20
Hairstyle changes fool FR 12% time (2022)
21
FR underperforms 18% on beards (2021)
22
Scar bias in FR 22% error increase (2020)
23
FR 30% worse on diverse lighting (2022)
Interpretation
Bias Statistics Interpretation
AI facial recognition technology, for all its vaunted promise, has been repeatedly shown to be staggeringly biased—with false positive rates of 2.3% for darker-skinned females (compared to 0.4% for lighter-skinned males), 34.7% higher gender classification errors for Black women than white men, up to 38% misgendering of trans individuals, 10x higher false match rates for Black faces, 5-10 year younger age estimates for dark skin tones, and far worse accuracy for Asian, Indigenous, and Indian faces—while also amplifying biases against women of color, glasses wearers, surgical mask users, and those with beards, scars, or diverse lighting, and performing poorly with children, the elderly, and underrepresented disabilities, revealing that "neutral" tech often mirrors and even magnifies human inequities.
03 · Category
Market Stats23 stats
01
Global facial recognition market size was valued at USD 4.0 billion in 2020 and expected to grow to USD 16.7 billion by 2028 at a CAGR of 19.4%
02
Facial recognition software market projected to reach USD 12.49 billion by 2026 from USD 4.91 billion in 2020, CAGR 16.7%
03
Asia-Pacific facial recognition market to grow at 23.5% CAGR to 2027
04
Investment in facial biometrics reached $1.2B in 2022 per PitchBook
05
Facial recognition revenue in China $2.3B in 2021
06
Global FR market CAGR 17.5% to $149B by 2030
07
US FR market $8.5B by 2025 forecast
08
Europe FR market to $15B by 2028, CAGR 22%
09
Biometrics market incl FR $95B by 2025
10
FR software patents grew 300% 2015-2020
11
Surveillance FR market $50B by 2030
12
Mobile FR market $25B by 2027
13
Cloud FR services market CAGR 25% to 2028
14
Retail FR market $7B by 2026
15
Healthcare FR $5B by 2028
16
Government FR spending $10B globally 2022
17
Automotive FR market CAGR 28% to 2030
18
Smart city FR investments $20B 2023-2028
19
Law enforcement FR $3B market 2025
20
Payment FR market $12B by 2027
21
Enterprise FR deployments up 40% YoY (2023)
22
Gaming FR market $2B by 2028
23
Border control FR $4B 2025
Interpretation
Market Stats Interpretation
Facial recognition is booming: valued at $4 billion in 2020, it’s projected to hit $149 billion by 2030 (with a 17.5% CAGR), spreading across surveillance, retail, healthcare, gaming, and beyond, while patents have tripled since 2015, enterprise deployments are up 40% in 2023, investment reached $1.2 billion in 2022, and from border control to smart cities, it’s no longer just cutting-edge—it’s practically everywhere you look.
More related reading
04 · Category
Privacy Incidents23 stats
01
Clearview AI scraped over 3 billion images from social media without consent by 2020, leading to privacy lawsuits
02
iBorderCtrl EU project rejected 47% of travelers based on facial analysis lies detection (2019 pilot)
03
San Francisco banned police use of facial recognition in 2019, first major US city
04
Meta sued for scanning 1B+ photos for facial data without consent (Illinois 2021)
05
Clearview AI data breach exposed 1M+ records in 2021
06
Delhi Police used FR to identify 3000+ suspects in 2022
07
UK's 20 live FR deployments scanned 560k faces in 2021
08
Russia FR system identified 200k+ violators in Moscow 2022
09
Facebook paused FR after 1B faces mapped (2021)
10
China 600M cameras with FR by 2021
11
EU fines company €20M for illegal FR use (2022)
12
1.4B faces in India's Aadhaar FR database (2023)
13
Brazil FR trial wrongfully arrested 7 innocents (2021)
14
Singapore scans 3M faces/day at borders (2023)
15
Detroit FR wrong ID led to arrest (2020)
16
London Met FR matched 1 in 1000 wrongly (Biometrics Inst 2020)
17
Nigeria FR database hacked, 60M IDs exposed (2023)
18
EU AI Act classifies FR as high-risk (2023)
19
US city pauses FR after 25% error rate (2022)
20
500k wrongful scans in one UK trial (2022)
21
Australia FR error arrested wrong man (2023)
22
Canada halts FR after privacy breach (2022)
23
100+ US police depts halt FR (2023)
Interpretation
Privacy Incidents Interpretation
Facial recognition technology, once a buzzworthy tool for innovation, has instead morphed into a global spectacle of privacy perils—with 3 billion images scraped, 1 million records exposed, and wrongful arrests ranging from seven innocents in Brazil to 3,000 suspects in Delhi—sparkling bans in San Francisco and 100+ U.S. cities, drawing €20 million fines in the EU, and spurring regulation as its errors, from 25% misidentifications to 1 in 1,000 wrong matches, have turned it into a tool that both fascinates and terrifies.
05 · Category
Usage Adoption23 stats
01
76% of consumers are uncomfortable with facial recognition in retail stores according to Deloitte survey (2022)
02
64% of US police departments use facial recognition as per Urban Institute survey (2021)
03
91% of Americans want federal regulation on facial recognition per Data & Society (2021)
04
30 million daily face scans by UK police ANPR cameras (2022 est.)
05
85% of Fortune 500 companies piloting facial recognition (2023 Gartner)
06
52% of EU citizens oppose FR in public spaces (Eurobarometer 2022)
07
70% of retailers using FR for theft prevention (NRF 2023)
08
62% US adults aware of FR, 56% concerned (Pew 2022)
09
Airports using FR for 100M+ passengers/year globally (ICAO 2023)
10
80% healthcare orgs adopting FR for patient ID (HIMSS 2023)
11
45% enterprises use FR for access control (IDC 2023)
12
55% schools considering FR for security (2023 survey)
13
68% banks using FR for KYC (2023)
14
75% stadiums deploy FR for entry (2023)
15
40% contactless payments via FR (Visa 2023)
16
90% airlines testing FR boarding (IATA 2023)
17
60% hotels adopt FR check-in (2023)
18
50% consumers accept FR for personalized ads (Kantar 2023)
19
65% events use FR ticketing (2023)
20
72% offices plan FR access (2024)
21
82% gyms use FR for members (2023)
22
58% transport hubs FR equipped (2023)
23
77% consumers trust FR in healthcare (Deloitte 2023)
Interpretation
Usage Adoption Interpretation
Amidst a landscape where 85% of Fortune 500 companies are testing facial recognition (FR), 100 million+ airport passengers are scanned yearly, and 80% of hospitals use it for patient IDs, the data also reveals a technology that’s hard to escape—yet 76% of consumers are uncomfortable with it in retail, 91% want federal regulation, and 64% of U.S. police rely on it, creating a dynamic where its reach often outpaces public comfort, even as 77% trust it in healthcare or 50% accept it for ads.
This sentence balances wit (via the "hard to escape" and "reach often outpaces public comfort" phrasing) with seriousness by framing the tension between ubiquity and unease, while weaving in key stats (adoption rates, regulatory desires, comfort levels) in a natural, human flow. It avoids lists or clunky structure, instead connecting trends through narrative contrast.
Reference
Cite This Report
This report is designed to be cited. We maintain stable URLs and versioned verification dates. Copy the format appropriate for your publication below.
APA
Gabrielle Fontaine. (2026, February 24). AI Facial Recognition Statistics. Gitnux. https://gitnux.org/ai-facial-recognition-statistics
MLA
Gabrielle Fontaine. "AI Facial Recognition Statistics." Gitnux, 24 Feb 2026, https://gitnux.org/ai-facial-recognition-statistics.
Chicago
Gabrielle Fontaine. 2026. "AI Facial Recognition Statistics." Gitnux. https://gitnux.org/ai-facial-recognition-statistics.
Sources & references
78 datasets cited across this report · attribution is report-level