3D:2D Kiosk POS - Matching Guide

Please read the Intro to 3D Face Matching before getting started with Matching.

FaceTec's 3D to 2D Kiosk POS Matching API is designed to Match 3D FaceMaps against Kiosk and POS style images. This API is useful for anyone creating their own front-end capture mechanism for a Kiosk/POS system.

The API is designed for use cases where a secure 3D Liveness Check can be done ahead of time and associated with a User, then in the future, a Kiosk/POS system can perform a 2D to 3D Match of the User to identify them.  After identification of the User, you will likely want to perform a 3D:3D 1:1 Re-Verification of that User to prove that they are present, not a Spoof, and for a higher Match Level (ML10+).

Performance

This API returns Match Levels up to 99.9998% (i.e. FARs up to 1 in 500,000 @ <1% FRR).

Get Started: 3D:2D Kiosk POS API

This API requires a 3D FaceMap that has Proven Liveness. Once you have created one of these, call the 3D to 2D Kiosk POS FaceTec Server API with the 3D FaceMap and the 2D Image. This API uses an externalDatabaseRefID to store and lookup an Enrolled User's 3D FaceMap.

Please Note: The externalDatabaseRefID is 100% controlled by your business logic and code. FaceTec does not set this internally (but we do demonstrate how to set it in our samples). Setting & Retrieving the externalDatabaseRefID should be done in Server-Side Code, not your Device-Side code.

Match Levels

For 3D:2D Kiosk POS Matching, Match Levels 0 through 7 are returned at 1% FRR.

  • Match Level 7 - 99.9998% (1/500,000 FAR)

  • Match Level 6 - 99.999% (1/100,000 FAR)

  • Match Level 5 - 99.99% (1/10,000 FAR)

  • Match Level 4 - 99.99% (1/1,000 FAR)

  • Match Level 3 - 99.8% (1/500 FAR)

  • Match Level 2 - 99.6% (1/250 FAR)

  • Match Level 1 - 99% (1/100 FAR)

  • Match Level 0 - Non-match


Factors Influencing Performance

This API will achieve above Match Levels at low FRRs with good input data. Here are some factors that can negatively impact performance:

  • Image alterations, filters, face morphing, compression, upscaling, re-saving.

  • Extreme pose variations.

  • Low resolution, blurry, or otherwise low quality images.

  • Matching against Users wearing sunglasses is not supported.

  • Group photos - if multiple faces are found, the highest Matching face result will be returned. Images with more than 4 faces will be rejected.

  • Other extreme conditions like extreme lighting, extreme facial expressions, or other extreme artistic alterations.

  • Homogenous datasets, matching with close relatives, and contrived testing with look-alikes.

2D Image Requirements & Recommendations

Your existing 2D Face Images can be used with FaceTec. However, the following set of requirements and recommendations for using 2D images must be followed. Your matching accuracy will vary depending on how far your image database deviates from the following requirements and recommendations:

  • JPEG or PNG. No other image formats are accepted.

  • Face has a frontal pose, mugshot style. Variations in pitch, roll, and yaw will lead to poor accuracy.

  • Face should be generally centered in the image. This is not a requirement, but it is helpful for performance.

  • Face width should be at least 150 pixels wide, ideally 300+ pixels wide. Smaller faces are sometimes accepted, but their use will degrade accuracy.

  • Images containing glare, watermarking, or holograms can negatively affect accuracy. 

  • Images should not be heavily compressed, and ideally not compressed more than once.

  • Face images should be upright. FaceTec SDKs will detect 90, 180, and 270-degree rotated faces in images, but processing will be faster if the photo is upright.

In general, conformance to ISO 19794-5 Frontal Face specifications will yield better performance. These include the following guidelines:

  • Neutral expression.

  • Good, even, ambient lighting.

  • Evenly lit face, minimizing any shadows and potential glare or reflections in glasses.

  • High-resolution image. A good baseline is around 640x360 portrait.

Factors Influencing Performance

This API will achieve the above Match Levels at low FRRs with good input data. Here are some factors that can negatively impact performance:

  • Image alterations, filters, face morphing, compression, upscaling, re-saving.

  • Extreme pose variations.

  • Low-resolution, blurry, or otherwise low-quality images.

  • Matching against Users wearing sunglasses is not supported.

  • Other extreme conditions like extreme lighting, extreme facial expressions, or other extreme artistic alterations.

  • Homogenous datasets, matching with close relatives, and contrived testing with look-alikes.

  • Glare or holograms obscuring or altering the face.

Lower Quality Images

Lower-Quality Images that do not meet the above guidelines, or have other quality-related aspects, should use the 3D:2D Kiosk POS Matching Low-Quality API. The Low-Quality Version of this API removes the minimum quality requirements for images, but can only achieve a maximum Match Level of 5.

Important Note: If the Default Version is unable to process a 2D Image, then FaceTec recommends using the Low-Quality Version of the 3D:2D Kiosk POS Matching Low-Quality API.