3D:3D Re-Verification

3D Face Matching is the ability for a computer system to estimate the probability that two 3D FaceMaps were derived from the same human. The FaceTec 3D Matching Algorithms are the most accurate in the world, with a FAR of 1 in 125,000,000 and a FRR less than 0.99%. By utilizing FaceTec's Patented Technology, the FaceTec 3D Matching Algorithms are able to operate on 3D Data captured from a 2D Camera to achieve Matches that are at least two Orders of Magnitude more accurate than those returned by any 2D Matching Algorithm.

FaceTec provides ALL Face Matching for FREE to all Customers & Partners. That includes 3D:3D, 3D:2D Portrait, 3D:2D Photo ID, 1:N for De-duplication, and even 2D:2D (upon request).

Introduction & Background

3D Liveness Detection & 3D Face Matching are the two primary functions of the FaceTec SDKs.

FaceTec 3D Liveness Detection is the only AI backed by a Level 1-5: 600K Spoof Bounty Program, has been certified to numerous standards, including NIST/NVLAP/ISO 30107, and is trusted to perform billions of 3D Liveness Checks annually.

FaceTec's 3D Matching Algorithms are the most accurate in the world. FaceTec leverages 3D FaceMap Technology to achieve levels of accuracy approximately 10 times better than Apple’s Face ID and 50-100 times better than the leading vendors in 2D matching, as confirmed by NIST FRVT testing.

Please see the FaceTec SDK 1/125,000,000 FAR Matching Report for information on the FaceTec 3D Matching Algorithms themselves, as well as for extensive comparisons of our algorithm accuracy against other industry benchmarks.

3D:3D Matching (1:1)

The first step to performing 3D:3D Matching is to create the original source 3D FaceMap by first performing a 3D Liveness Check on the desired User. The 3D FaceMap generated upon Successfully Proving 3D Liveness is then stored in Your Database.

Next, when the User needs to be Re-Verified, you perform a 3D Liveness Check once again, followed by a 3D:3D Match. 3D:3D Matching can only succeed if and only if first 3D Liveness is Proven on the incoming 3D FaceScan, then if and only if the resulting 3D FaceMap Matches the Enrollment 3D FaceMap at >1/125,000,000 FAR (Match Level 15).

When using the FaceTec Server HTTPS API Endpoints, 3D:3D Matching uses an externalDatabaseRefID to store and look up an Enrolled User's 3D FaceMap.

Please Note: The externalDatabaseRefID is 100% controlled by Your Code.  FaceTec does not set this internally or via any hidden logic. FaceTec's Sample Apps demonstrate setting this value, but in Your App and Your Code, you need to configure this per your own business logic. Setting/Retrieving the externalDatabaseRefID should be done in Server-Side Code, not your Device-Side code.

Please note that 1:1 3D:3D Matching does not search across all enrolled users.

 

Match Levels

FaceTec Match Levels & associated probabilities at <0.99% FRR:

  • Match Level 15 - 1/125,000,000 FAR

  • Match Level 14 - 1/95,000,000 FAR

  • Match Level 13 - 1/70,00,000 FAR

  • Match Level 12 - 1/50,000,000 FAR

  • Match Level 11 - 1/25,000,000 FAR

  • Match Level 10 - 1/12,800,000 FAR

  • Match Level 9 - 1/2,000,000 FAR

  • Match Level 8 - 1/1,000,000 FAR

  • Match Level 7 - 1/500,000 FAR

  • Match Level 6 - 1/100,000 FAR

  • Match Level 5 - 1/10,000 FAR

  • Match Level 4 - 1/1,000 FAR

  • Match Level 3 - 1/500 FAR

  • Match Level 2 - 1/250 FAR

  • Match Level 1 - 1/100 FAR

  • Match Level 0 - Non-match

Adaptive 3D FaceMaps

During Enrollment, a 3D FaceMap will be generated once the User Successfully Proves 3D Liveness, then stored in Your Database for future 3D:3D Re-Verifications. To succeed in a 3D:3D Re-Verification, the User must first Prove 3D Liveness and Match the corresponding Enrolled Liveness-proven 3D FaceMap at Match Level 15.

In the Real-World, new 3D:3D Re-Verifications are performed under different conditions (changes in lighting, camera angles, facial expressions, or minor appearance variations) compared to when the User was first Enrolled in Your Application. To improve Usability and User Experience in these scenarios, FaceTec developed Adaptive 3D FaceMaps.

Adaptive 3D FaceMaps are enabled by default when using one of the standard database implementations. Adaptive 3D FaceMaps provide an improved 3D FaceMap with every successful 3D:3D Re-Verification, enhancing the Usability of the 3D:3D Re-Verification process over time.

1:N Search

3D:3D Matching can be performed against a 3D Database of Users for use cases like detecting Duplicate Accounts or maintaining a list of Fraudulent Users.

To learn more about 1:N Search, see 1:N Search for De-duplication & Fraud Lists (ABIS).

FaceTec 3D FaceMaps Information

The 3D FaceMap technology is relied upon by all core functions of the FaceTec SDK platform.

FaceTec 3D FaceMaps are generated from every completed User Session. Additionally, FaceTec 3D FaceMaps are all of the following:

  • Encrypted: This is required and enforced by the FaceTec SDK APIs.

  • Proprietary: FaceTec 3D FaceMaps cannot be used by anything other than the FaceTec Server.

  • Created via a patented process: Users performing User Sessions from any device are using FaceTec’s patented interface to create FaceTec 3D FaceMaps.

  • Cross-platform: 3D FaceMaps captured on iOS, Android, or Browser can be matched interchangeably.

  • Data-rich: 3D FaceScans/Maps are created by processing upwards of 100 frames of video.

  • Future-proof & forward-compatible: FaceTec 3D FaceMaps created today are guaranteed to be able to leverage future FaceTec Matching Algorithm improvements.

  • Not a Honeypot Risk: Liveness Data used for Liveness Detection is deleted from FaceMaps after each session.

  • Flexible: FaceTec 3D FaceMaps can be stored as files, in databases, in blocks, on a server, or on a device. Customer-specific encryption schemes can be applied in addition to FaceTec SDK built-in security mechanisms.

Why is 3D Face Matching Important?

3D FaceMaps are much more accurate replicas of the 3D Human users they were collected from than 2D photos or videos are.  This increase is in both depth and breadth of the data, and that data enables FaceTec's AI to recreate any frontal 2D face photo scenario and match to another 3D FaceMap with astounding accuracy and without bias.  For biometric security, high-confidence Face Matching is important for Onboarding, Legal Identity Verification, Re-verification, and Anti-Fraud use cases.

FaceTec Matching APIs

Real-World Security Metrics

When biometrics salesmen tout accuracy numbers like "99.9% Accurate" — In reality, that is horribly inaccurate, and you are being tricked by your own human bias.  Intuitively, you think 99% is pretty good, so 99.9%, must be GREAT, but it's not.  You need to start thinking in terms of 99.999%, 99.9999% accurate, etc.

Did your eyes glaze over when you read numbers like 99.999%, and 99.9999%?  These accuracy numbers are hard to process for human brains, but when you have hundreds of thousands, or millions of users, which we hope you do, these accuracy percentages become very important.

Is FaceTec 3D:3D Really 100x More Accurate than 2D:2D?

A FaceTec SDK Session processes many hundreds of images directly from the User's Camera.  The AI then extracts 3D Data from the sequential 2D frames, measures the perspective distortion observed, and renders a 3D FaceMap of the user.  2D matching algorithms like those in the NIST tests use one 2D Image and match it to another 2D image.  Depending on the capture distance, the camera lens, the height of the camera vs. the subject, and other factors, the variations can overwhelm the 2D face matching AI, and the accuracy plummets.

If we look at the NIST "Authentication" Dataset — This is the MUGSHOT Dataset (Subjects All from the United States) — The best FAR reported on the NIST Leaderboard is 1/1,000,000.

Meanwhile, FaceTec's Dataset, which includes Data from over 170 Countries, achieves a FAR of 1/125,000,000.

Our internal testing performs Billions of N:N imposter comparisons to generate these FAR numbers. The number of errors generated by our Algorithm, over Billions of comparisons, is between 10 & 100.  The amount of errors generated by NIST Algorithms at 1/1,000,000 FAR is therefore on the order of TENS OF THOUSANDS.

Market Leading AI

FaceTec’s 3D:3D FAR is 1/125,000,000. This means that it would take 125,000,000 random imposter attempts before you would expect to see 1 error. FaceTec achieves this accuracy under Real-World usage, and the measurement is taken across a Dataset of volunteer Users from 170+ Countries and Billions of matching attempts.

Honorable Mention: Apple’s Face ID has FAR of 1/1,000,000 (See iPhone X Keynote), but is not usable for Real Authentication because it can be bypassed with a PIN, and it doesn't work across devices, platforms, or the web, and you cannot compare Face ID scans with Photo IDs or On-File Trusted Photos.

3D Makes All The Difference

What do the above two companies that have a Security Rating of 1 and Face Matching technology have in common?

Answer: They both use 3D Face Matching Data.

What is 1/125,000,000 FAR?

The larger the denominator, the lower the chance that a human imposter can impersonate one of your other users.

We realize that 1/125,000,000 and 1/100,000 are both small decimal numbers; however, what matters is how these numbers play out at scale.
 

A Footnote For Biometrics Scientists

  • The Security Rating directly maps to FAR/TRR/FNMR. It is exactly the same metric, but repackaged so that Real-World security conclusions can be drawn from it. Technically speaking: Security Rating = (1/FAR) —> Expressed in Whole Number increments of 100,000 because in the Real World, there are Whole Numbers of Real Users.
  • Companies selling accuracy as “99.99%” as a “good thing” are either ignorant or openly corrupt. 99.99% == 1/10,000 == Security Rating of 10,000 == Pervasive Security Risk.

FaceTec Matching APIs

FaceTec has Matching APIs tailored to every Real-World Matching Use Case (outside of 2D:2D surveillance). FaceTec doesn't do surveillance.

 

3D Liveness + 3D Matching (3D to 3D) - Re-Verification, Password Replacement

Notes: 3D Liveness + 3D Matching is also called Authentication or Re-Verification, depending on what is being matched against.

Security: 1/125,000,000 FAR

Read more at 3D:3D Matching - 1:1 Verification.

 

3D Liveness + Photo ID Match (3D to 2D) - Onboarding & KYC (eKYC Companies)

Notes: Uses FaceTec's Proprietary 3D FaceScan and ID Scan interfaces to Verify a user against their legal Photo ID

Security: 1/500,000 FAR

Read more at 3D FaceScan to Photo ID Match

 

3D Liveness + Face Portrait Match (3D to 2D) - Onboarding & KYC (eKYC Companies & Governments)

Notes: Uses FaceTec's Face Scan Interface and Matches to Original Images (not User Scans) of Users.

Security: 1/1,000,000 FAR

Read more at Identity Issuers = Digital Verifiers

 

3D Liveness + Matching to 2D ID Photos from 3rd Party Sources

Notes: If you are already using another provider to scan User ID Photos, you can compare these ID Photos to the 3D FaceMap at high confidence levels.

Security: 1/500,000 FAR

Read more at 3D:2D 3rd-Party ID Photo Image

 

3D Liveness + Dating & Social Media Profile Pics

Notes: For ensuring that the User's 3D FaceMap Matches the face images the User Uploads of themselves.

Security: 1/100,000 FAR

Read more at 3D:2D Social Media Profile Pic - Matching Guide