Selfie & Face Match
Available in SDK v1.0+
The Selfie & Face Match capability verifies that the person completing the workflow is physically present and matches the photo on a previously scanned identity document. It produces two independent scores: a face match score and a liveness score. Both scores can be used independently to define rules and outcomes.
This capability must come after a Scan ID or Scan & Verify ID step, since it compares the selfie against the face image extracted from the document.
How it works
When the user reaches this step, the SDK captures a short selfie video clip. From that clip, the capability:
- Checks for liveness: determines whether the person is physically present, as opposed to someone holding up a photo or playing a video of another person.
- Compares faces: computes the probability that the face in the selfie is the same person as the face on the identity document captured in the previous step.
Both checks use passive AI-based analysis; the user doesn't need to perform any specific action like blinking or turning their head.
Configuration
You can configure the face match tolerance, which determines how well the selfie must match the ID photo. There are three levels:
- Soft: tolerates more variation between the selfie and the ID photo; lower false rejection rate but higher false acceptance rate
- Regular: a balanced middle ground; recommended for most use cases
- Hard: less tolerant of differences; lower false acceptance rate but higher false rejection rate
The tolerance level affects the face match score returned in the response. The same underlying faces at the same threshold will yield a higher score at "soft" and a lower score at "hard." This means that if you set the same numeric threshold in your rules, a "soft" calibration will accept more borderline matches than "hard."
Rules
Rules can be configured along two independent dimensions:
-
Face match score: the probability (0 to 100) that the person in the selfie is the same person shown on the ID document. If this score falls below the threshold you specify, the user is rejected or sent to manual review.
-
Liveness score: the probability (0 to 100) that the person in the selfie is physically present. If this score falls below the threshold you specify, the user is rejected or sent to manual review.
Because the face match score is influenced by the tolerance calibration, the same numeric threshold produces different effective strictness depending on which tolerance level you selected. Choose your threshold after deciding on the tolerance level, and test the combination against your user population to find the right balance.