Face Match & Likeness

Face checks answer: is this the right person — and does it stay the right person?

Three related capabilities work together:

1:1 Match (likeness)

Compares two faces — typically the live selfie or video frame against the photo on the submitted ID — and returns a likeness score.

{
  "check": "face_match",
  "verdict": "pass",
  "score": 0.91,
  "reference": "kyc_pan_card.photo",
  "probe": "session.selfie"
}

The comparison is robust to ageing, glasses, lighting, and print quality of the ID photo. Thresholds are configurable per agent — regulated flows commonly gate at 0.85+.

1:N Search

Searches a face against your configured biometric databases: watchlists, prior applicants, known-fraud galleries. A hit returns the matched identity, the database it came from, and the match confidence — the raw material for duplicate-application and repeat-fraudster detection.

See the Face Biometrics API for endpoints (/face/match, /face/search, /face/register).

Face anchor (session continuity)

In a video session, matching the face once at the start is not enough — the genuine customer can pass the opening check and then hand the call to someone else. The face anchor locks onto the verified face and confirms the same person remains on camera throughout, flagging:

• Person swaps mid-session
• Multiple faces appearing when only the subject should be present
• Extended absences from frame during critical steps

Anchor breaks are reported as timestamped markers in the session timeline.

Gating with Goal Rules

[
  {
    "id": "selfie_matches_id",
    "severity": "hard_stop",
    "check": "face_match_gte",
    "params": { "threshold": 0.85, "reference": "kyc_pan_card.photo" },
    "on_deny": { "message": "Selfie does not match ID photo.", "max_retries": 0, "remediate_type": "" }
  },
  {
    "id": "single_subject_throughout",
    "severity": "require",
    "check": "face_anchor_unbroken",
    "params": {},
    "on_deny": { "message": "Face anchor broke mid-session — route to review.", "max_retries": 0, "remediate_type": "" }
  }
]

Related checks

• Liveness — likeness without liveness can be spoofed with a photo of the victim.
• Deepfake Detection — likeness without deepfake detection can be spoofed with a face swap of the victim.
• Cross-Verification — the photo on every submitted ID is also cross-matched.