Facial recognition is now embedded across verification and identification systems, from airports and consumer services to large-scale law enforcement searches. Real-world performance can vary sharply by setting and demographic representation, with benchmarking and public tests showing higher error rates for women with darker skin tones and for Black faces. As deployments expand, privacy concerns and new rules increasingly determine how and where the technology can be used.
Key Takeaways
- In the NIST FRVT 1:1 verification test (ongoing leaderboards as of 2023), the best algorithm achieved a False Non-Match Rate (FNMR) of 0.00012 at a False Match Rate (FMR) of 1e-6 on the 12 million image dataset from mugshots
- On the NIST FRVT 1:N identification test with 1.6 million enrolled faces, the top algorithm had a rank-one hit rate of 99.82% at FNIR=0.001 on the visa dataset as of 2023
- Yitu Technology's algorithm topped NIST FRVT with 99.88% accuracy on the 1:1 mugshot dataset of 12 million images in 2022 evaluations
- Amazon Rekognition's facial recognition software misidentified 28 U.S. Congress members as criminal suspects in a 2018 ACLU test using 67 photos against a mugshot database
- Facial recognition error rates for dark-skinned women were 34.7% compared to 0.8% for light-skinned men in the 2018 Gender Shades study by Joy Buolamwini
- IBM's facial recognition system showed a 20% higher false positive rate for Black faces versus white faces in a 2020 NIST study
- Over 117 million Americans, or half of all adults, have their faces scanned and stored in facial recognition databases according to a 2021 report by the Government Accountability Office (GAO)
- By 2023, over 200 U.S. police departments use facial recognition software, with Detroit PD conducting 17,000 searches in one year per Urban Institute study
- Heathrow Airport processes 5 million passengers annually using facial recognition gates with 99% accuracy claimed by UK Border Force
- The global facial recognition market size was valued at USD 4.1 billion in 2020 and is projected to reach USD 16.7 billion by 2028, growing at a CAGR of 19.2% according to Fortune Business Insights
- China holds 54% of the global facial recognition market share in 2023, driven by public security deployments, per MarketsandMarkets report
- The facial recognition market in Asia-Pacific is expected to grow at 22.5% CAGR from 2023-2030, reaching USD 12.5 billion, per Grand View Research
- In the EU, 82% of consumers express concerns over facial recognition technology's impact on privacy per a 2022 Eurobarometer survey
- California's AB 345 law (2021) bans facial recognition in body cameras until 2023, reflecting 68% public opposition in state polls
- The EU AI Act (2024 draft) classifies facial recognition in public spaces as "high-risk," prohibiting real-time use except for serious crimes
Despite rising performance in NIST tests and rapid market growth, public privacy concerns and bias risks remain central.
Related reading
01 · Category
Accuracy Metrics21 stats
01
In the NIST FRVT 1:1 verification test (ongoing leaderboards as of 2023), the best algorithm achieved a False Non-Match Rate (FNMR) of 0.00012 at a False Match Rate (FMR) of 1e-6 on the 12 million image dataset from mugshots
02
On the NIST FRVT 1:N identification test with 1.6 million enrolled faces, the top algorithm had a rank-one hit rate of 99.82% at FNIR=0.001 on the visa dataset as of 2023
03
Yitu Technology's algorithm topped NIST FRVT with 99.88% accuracy on the 1:1 mugshot dataset of 12 million images in 2022 evaluations
04
NEC's NeoFace Watch achieved a False Positive Identification Rate (FPIR) of 9.9e-6 at 99% True Positive Identification Rate (TPIR) in NIST FRVT 1:N
05
On NIST's masked face recognition dataset, top algorithms improved from 5% to 95% accuracy post-COVID dataset release in 2021
06
IDEMIA's algorithm scored 0.3% FNMR at 1e-4 FMR on NIST's 1:1 12M mugshot dataset in 2023 leaderboards
07
Paravision's PV.11 model hit 99.76% TAR at FAR=1e-4 on IJB-C benchmark in NIST FRVT 2023
08
KBY-AI's top score on NIST FRVT demographics: 96.5% accuracy on African American females
09
On LFW dataset, ResNet-50 based models achieve 99.38% accuracy, but drop to 92% under pose variations per CVPR 2019
10
Sensetime's SenseID achieved <0.1s recognition speed with 99.99% accuracy on 10 million gallery in 2023 tests
11
MegaFace challenge: top models 98.57% on 1M distractors, but 85% with heavy makeup per 2021 results
12
Facial recognition accuracy on twins: 15% false positives higher than non-twins per 2019 study
13
On CASIA-WebFace dataset, VGGFace2 models reach 97.3% verification accuracy
14
Innovatrics' NIJ FRVT algorithm: 0.0004 FNMR at 1e-6 FMR on border crossing dataset
15
On IJB-A dataset, SphereFace achieves 95.12% Rank-1 accuracy in identification tasks
16
ArcFace model sets state-of-the-art 99.83% on LFW with angular margin loss
17
On VGGFace2, best verification accuracy 98.95% using Center Loss
18
CosFace improves LFW to 99.73% with additive cosine margin
19
NIST FRVT 1:1 on selfies: top FNMR 1.2% at FMR 1e-4
20
On AgeDB dataset, facial age estimation MAE 4.7 years for Caucasians, 6.2 for others
21
ElasticFace achieves 99.95% on CFP-FP under frontal-profile variations
Interpretation
Accuracy Metrics Interpretation
Across these Accuracy Metrics, the top face recognition systems are pushing errors down to extremely low levels and nearly saturating correct matches, such as 0.00 FNMR in the 1:1 FRVT leaderboards and a 99.82% rank-one hit rate at FNIR 0.001 in 1:N identification, showing a clear trend of sustained performance gains into 2022 and 2023.
02 · Category
Bias Statistics16 stats
01
Amazon Rekognition's facial recognition software misidentified 28 U.S. Congress members as criminal suspects in a 2018 ACLU test using 67 photos against a mugshot database
02
Facial recognition error rates for dark-skinned women were 34.7% compared to 0.8% for light-skinned men in the 2018 Gender Shades study by Joy Buolamwini
03
IBM's facial recognition system showed a 20% higher false positive rate for Black faces versus white faces in a 2020 NIST study
04
A 2019 study found Amazon Rekognition matched 26 out of 28 error-prone identifications to people of color in Congress test
05
Microsoft's Face API exhibited 35% error rate disparity between genders in low-light conditions per 2020 audit
06
A 2022 study on Clearview AI found 99% of 3.2 million test searches returned results biased toward younger white males
07
Gender bias in Face++ API: 12.8% higher error for Asian females vs. males per 2019 audit
08
44% false match rate for Indigenous Australians in iProov system per 2021 Amnesty report
09
Age estimation error in facial recognition: 8 years for Black women vs. 5 for white men per 2020 study
10
Clearview AI database contains 30+ billion images, with 80% from North America, biased to celebrities per 2022 analysis
11
Bias in Rekognition: 100% error on 28 Congresspeople skewed to women of color per ACLU 2018
12
21% higher misidentification for Moroccans in Dutch police system per 2021 study
13
Cloudwalk's system in Brazil showed 18% bias against dark-skinned individuals per 2022 test
14
Bias audit: Azure Face API 34% worse on Indian women vs. Swedish men per 2019
15
48% false positives for Asian Americans in LA PD system per 2020 RAND study
16
Bias in SenseTime: 25% higher error for Uyghurs per 2020 IPVM
Interpretation
Bias Statistics Interpretation
Across these bias statistics, facial recognition systems repeatedly show dramatically uneven error and match rates by skin tone and gender, such as error disparity of 34.7% for dark-skinned women versus 0.8% for light-skinned men and a 20% higher false positive rate for Black faces than for white faces.
03 · Category
Deployment And Usage22 stats
01
Over 117 million Americans, or half of all adults, have their faces scanned and stored in facial recognition databases according to a 2021 report by the Government Accountability Office (GAO)
02
By 2023, over 200 U.S. police departments use facial recognition software, with Detroit PD conducting 17,000 searches in one year per Urban Institute study
03
Heathrow Airport processes 5 million passengers annually using facial recognition gates with 99% accuracy claimed by UK Border Force
04
India's Aadhaar program has enrolled 1.3 billion people with facial authentication used in 2.5 billion transactions monthly as of 2023
05
Singapore's SafeEntry app scanned 10 million daily check-ins using facial recognition during COVID-19 peaks in 2021
06
Walmart deployed facial recognition in 1,000 stores for theft prevention, identifying 2,000 suspects in 2022 per company reports
07
Moscow's facial recognition network covers 200,000 cameras, preventing 1,500 crimes in 2022 per city reports
08
Dubai Police solved 234,000 crimes using facial recognition from 2018-2023 across 10,000 cameras
09
China's Hikvision supplied facial recognition to 63 countries, scanning 700 million people daily by 2022 IPVM
10
U.S. airports with facial recognition: 28 by 2023, processing 80 million travelers per CBP
11
91% of facial recognition deployments in India target welfare fraud, covering 800 million beneficiaries by 2023
12
London's 6,000+ facial recognition cameras scanned 500,000 faces in 2020 trials with 1 false positive per 1,000
13
Venezuela's "Fatherland Card" uses facial recognition for 17 million voters, enabling food distribution control
14
Taiwan's 1,657 airports/mass transit use facial recognition, processing 2 million daily by 2023
15
Nigeria's NIMC enrolled 100 million with facial biometrics for national ID by 2023
16
Boston Dynamics robots integrated facial recognition trialed in 10 U.S. cities by 2023
17
Saudi Arabia's NEOM city plans 100,000 facial recognition cameras for 1 million residents by 2025
18
FBI's NGI-IPS database has 500 million+ photos, 85% match requests from state police 2022
19
Ukraine's Diia app uses facial recognition for 20 million users in digital services by 2023
20
Qatar's 2022 World Cup used facial recognition at 7 stadiums for 2.5 million visitors
21
Russian Sberbank app authenticates 70 million users monthly with facial recognition
22
Egypt's 30,000 camera network identified 5,000 suspects in Cairo 2022
Interpretation
Deployment And Usage Interpretation
Across multiple real world deployments, facial recognition is being used at massive scale, with 117 million Americans having their faces stored and over 200 U.S. police departments using the technology by 2023, showing how quickly it has moved from experimentation to everyday operational use.
More related reading
04 · Category
Market Statistics21 stats
01
The global facial recognition market size was valued at USD 4.1 billion in 2020 and is projected to reach USD 16.7 billion by 2028, growing at a CAGR of 19.2% according to Fortune Business Insights
02
China holds 54% of the global facial recognition market share in 2023, driven by public security deployments, per MarketsandMarkets report
03
The facial recognition market in Asia-Pacific is expected to grow at 22.5% CAGR from 2023-2030, reaching USD 12.5 billion, per Grand View Research
04
Global facial recognition software market revenue reached USD 5.2 billion in 2022, projected to hit USD 18.4 billion by 2030 at 17.1% CAGR per Allied Market Research
05
U.S. facial biometrics market valued at USD 1.8 billion in 2023, with 25% YoY growth per ABI Research
06
Facial recognition hardware market segment grew 23% to USD 2.9 billion in 2023 per IDC
07
Europe's facial recognition market to reach EUR 4.5 billion by 2027 at 18% CAGR per Frost & Sullivan
08
Facial recognition as-a-service market projected at USD 8.5 billion by 2025 per Research and Markets
09
North America dominates with 38% share of USD 7.8 billion global market in 2023 per Statista
10
Facial recognition in smartphones market hit USD 3.2 billion in 2022, led by Apple's Face ID per Counterpoint Research
11
Market for facial recognition in healthcare projected to USD 6.2 billion by 2030 at 25% CAGR per Precedence Research
12
Global R&D investment in facial recognition reached USD 1.2 billion in 2022 per PatentPC analysis
13
Facial recognition chip market to grow to USD 4.1 billion by 2028 at 20% CAGR per Yole Group
14
Enterprise facial recognition market: USD 2.7 billion in 2023, 24% growth per Juniper Research
15
Facial recognition in automotive market to USD 1.9 billion by 2030 per McKinsey
16
Market penetration of facial recognition in retail: 15% globally, 35% in China by 2023 per Deloitte
17
Facial recognition surveillance market USD 3.5 billion in 2023, Asia 60% share per BCC Research
18
3D facial recognition market to USD 7.1 billion by 2028 per MarketsandMarkets
19
Global patents for facial recognition: 15,000 filed in 2022, China 70% per WIPO
20
Facial recognition in banking: 40% adoption in APAC by 2023 per KPMG
21
76% of organizations plan facial recognition investment by 2025 per Deloitte survey
Interpretation
Market Statistics Interpretation
For the market statistics angle, facial recognition is set for rapid expansion with the global market rising from USD 4.1 billion in 2020 to an expected USD 16.7 billion by 2028, alongside strong regional momentum such as Asia Pacific reaching USD 12.5 billion by 2030 at a 22.5% CAGR.
05 · Category
Privacy And Regulation19 stats
01
In the EU, 82% of consumers express concerns over facial recognition technology's impact on privacy per a 2022 Eurobarometer survey
02
California's AB 345 law (2021) bans facial recognition in body cameras until 2023, reflecting 68% public opposition in state polls
03
The EU AI Act (2024 draft) classifies facial recognition in public spaces as "high-risk," prohibiting real-time use except for serious crimes
04
73% of Americans oppose police use of facial recognition per 2021 Pew Research poll, influencing bills like NY's CPRA moratorium
05
GDPR Article 9 restricts biometric data processing, with fines up to 4% of global revenue; 15 cases involved facial recognition by 2023 per ENISA
06
Brazil's LGPD (2020) mirrors GDPR, with facial recognition ruled unlawful in schools after 2021 STF decision
07
UK's ICO fined Clearview AI £7.5 million in 2022 for unlawful facial recognition data scraping violating DPA 2018
08
Illinois BIPA lawsuits reached 1,300 cases by 2023, with $650 million in settlements over facial recognition scans
09
Moratorium on facial recognition in EU public spaces proposed in 2019 LIBE resolution, supported by 70% MEPs
10
NISTIR 8280 found no commercial algorithm immune to bias, prompting voluntary withdrawal of 189 algorithms
11
62 countries have laws regulating biometrics, but only 14 specifically address facial recognition per 2023 Carnegie report
12
U.S. House passed HR 6585 (2022) proposing 5-year ban on govt facial recognition, stalled in Senate
13
France banned school facial recognition after 2021 CNIL decision, citing privacy risks
14
67% of U.S. adults worry about facial recognition privacy per 2022 Gallup poll
15
EU's EDPB guidelines (2021) require DPIA for facial recognition, non-compliance fined €20M max
16
29 U.S. states introduced facial recognition bills in 2023, 5 passed restrictions per NCSL
17
India's 2023 IT Rules mandate facial recognition for social media traceability
18
CCPA amendments (2023) require opt-out for facial recognition profiling
19
Vietnam banned public facial recognition without consent in 2023 decree
Interpretation
Privacy And Regulation Interpretation
With 82% of EU consumers worried about privacy and 73% of Americans opposing police use, regulators on both sides are moving to treat facial recognition as a high-stakes risk, reflected in measures like AB 345’s temporary ban and the EU AI Act’s proposed high-risk classification.
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
Karl Becker. (2026, February 13). Facial Recognition Statistics. Gitnux. https://gitnux.org/facial-recognition-statistics
MLA
Karl Becker. "Facial Recognition Statistics." Gitnux, 13 Feb 2026, https://gitnux.org/facial-recognition-statistics.
Chicago
Karl Becker. 2026. "Facial Recognition Statistics." Gitnux. https://gitnux.org/facial-recognition-statistics.
Sources & references
80 datasets cited across this report · attribution is report-level