Linking & Identification · Company practice
AI checks facial biometrics against the ID document
The AI indicates whether the facial biometrics supplied by an end user appear genuine and correspond to the individual shown on the identity document.
Details
- Who
- Entrust IDV AI
- What it does
- indicate genuineness and correspondence with the document portrait
- What it covers
- end-user facial biometrics
Supporting quotations
Indicate whether the facial biometrics provided by an end user appear genuine and corresponds to the individual shown on the identity document.
Qualifications & exceptions
- Framed as an indicator of correspondence, not proof of identity.
Not stated here
- The notice does not state the match threshold or how look-alike or presentation attacks are handled beyond liveness detection.
Linking & Identification · Company practice
Embedding-based 1:1 or 1:N face comparison
Entrust uses supervised deep-learning face-recognition models to compare a live-captured selfie or video with an image from a government-issued ID, or with other previously captured selfies or videos. The model produces an embedding for each face image and measures similarity to estimate the likelihood that both images belong to the same person; the notice calls this embedding-based, 1:1 or 1:N biometric verification.
Details
- Who
- Entrust IDV face-recognition models
- What it does
- compare face embeddings and estimate whether images show the same person
- What it covers
- live selfie/video, ID portrait and previously captured selfies/videos
Supporting quotations
Our application employs supervised deep learning models to perform facial comparisons between live-captured selfie or video and an image sourced from a government-issued identity document, or other previously captured selfies or videos.
The model produces a numerical representation (“embedding”) of each face image and measures the similarity between embeddings to determine the likelihood that both images belong to the same individual, otherwise known as an embedding-based, 1:1 or 1:N biometric verification.
Qualifications & exceptions
- The notice describes a likelihood estimate; it does not state match thresholds or error rates.
Not stated here
- What stored images or gallery support 1:N comparisons, and how long they are retained, is not explained in this notice.