Face Detection, Verification and Liveness

This content was generated with the assistance of AI and should be verified against the official documentation before being relied on in production.

See the section bibliography for the reference material consulted while preparing these pages.

This page targets Azure CLI 2.90.0 and the Face REST API v1.2. The Face client libraries are still preview: azure-ai-vision-face 1.0.0b2 (Python), Azure.AI.Vision.Face 1.0.0-beta.2 (.NET) and com.azure:azure-ai-vision-face 1.0.0-beta.2 (Java), checked on 2026-10-06. There is no GA Java SDK, so this page shows Python, C# and REST.

Azure Face in Foundry Tools (part of Azure Vision) detects human faces in images and, for approved customers, compares them. Face is the most restricted tool in the catalog. Read the Limited Access rules before you design anything around it.

Bookshelf doesn’t use Face

Bookshelf deliberately does not use Face. A bookshop has no need to identify or verify people from images, and adding biometric processing would bring consent, retention and legal obligations with no benefit to readers. The examples on this page use a generic KYC (know your customer) onboarding scenario for a regulated business instead.

Operations

Operation	What it does
Detect	Finds faces and returns a bounding box, landmarks, head pose, blur, exposure, occlusion, glasses, mask and a quality score for recognition. Available to everyone.
Verify (1:1)	Do two faces belong to the same person? Limited Access.
Identify (1:N)	Which person in a person group (or large person group) is this? Limited Access.
Find similar	Faces that look like a query face, in a face list. Limited Access.
Group	Clusters unknown faces by similarity. Limited Access.
Liveness detection	Is this a live person in front of the camera, not a photo, screen or mask? Uses a client SDK (iOS, Android, web) plus a server-side session. Liveness with verification also compares the live face to a reference photo.

Operation

What it does

Detect

Finds faces and returns a bounding box, landmarks, head pose, blur, exposure, occlusion, glasses, mask and a quality score for recognition. Available to everyone.

Verify (1:1)

Do two faces belong to the same person? Limited Access.

Identify (1:N)

Which person in a person group (or large person group) is this? Limited Access.

Find similar

Faces that look like a query face, in a face list. Limited Access.

Group

Clusters unknown faces by similarity. Limited Access.

Liveness detection

Is this a live person in front of the camera, not a photo, screen or mask? Uses a client SDK (iOS, Android, web) plus a server-side session. Liveness with verification also compares the live face to a reference photo.

Source: What is Azure Face?

Limited Access, attributes and prohibited uses

Limited Access: identification, verification, find similar, group and liveness with verification are available only to Microsoft-managed customers and partners whose use case is approved through the Face intake form.
Retired attributes: emotion, gender, age, smile, facial hair, hair and makeup were removed from detection.
Limited attributes: glasses, mask, head pose, blur, exposure, noise and occlusion remain, mainly to judge image quality.
Prohibited uses include use by or for a police department in the United States for facial recognition, and real-time identification in uncontrolled public spaces.
Input limits: JPEG, PNG, GIF or BMP; up to 6 MB; faces at least 36 x 36 pixels (200 x 200 for best results); up to 100 faces per image; 1 million persons per large person group.

Source: Limited Access features

Provisioning

In the Azure portal

Create a single-service Face resource (or use the Foundry resource) in a supported region.
Submit the Face Limited Access intake form, linked from the resource’s overview, before using verification, identification or liveness with verification.
In the Foundry portal, Playgrounds → Vision playground → Face tries detection and liveness.
Assign Cognitive Services Face Recognizer to the identity that calls Face.

With the Azure CLI

az cognitiveservices account create \
  --name face-kyc-demo --resource-group rg-kyc-demo --location eastus \
  --kind Face --sku S0 --custom-domain face-kyc-demo --yes

Source: Azure CLI reference — az cognitiveservices account create

Detection

from azure.ai.vision.face import FaceClient
from azure.ai.vision.face.models import (
    FaceDetectionModel, FaceRecognitionModel, FaceAttributeTypeDetection03,
)
from azure.identity import DefaultAzureCredential

client = FaceClient("https://face-kyc-demo.cognitiveservices.azure.com/", DefaultAzureCredential())
with open("id-photo.jpg", "rb") as f:
    faces = client.detect(
        f.read(),
        detection_model=FaceDetectionModel.DETECTION03,
        recognition_model=FaceRecognitionModel.RECOGNITION04,
        return_face_id=False,
        return_face_attributes=[FaceAttributeTypeDetection03.QUALITY_FOR_RECOGNITION, FaceAttributeTypeDetection03.BLUR],
    )
for face in faces:
    print(face.face_rectangle, face.face_attributes.quality_for_recognition)

Source: Face quickstart (Python)

using Azure.AI.Vision.Face;
using Azure.Identity;

var client = new FaceClient(
    new Uri("https://face-kyc-demo.cognitiveservices.azure.com/"), new DefaultAzureCredential());

var response = await client.DetectAsync(
    BinaryData.FromBytes(await File.ReadAllBytesAsync("id-photo.jpg")),
    FaceDetectionModel.Detection03, FaceRecognitionModel.Recognition04,
    returnFaceId: false,
    returnFaceAttributes: new[] { FaceAttributeType.QualityForRecognition });

foreach (FaceDetectionResult face in response.Value)
    Console.WriteLine($"{face.FaceRectangle.Width}x{face.FaceRectangle.Height} {face.FaceAttributes.QualityForRecognition}");

Source: Face quickstart (C#)

KYC onboarding with liveness

A bank’s app onboards a customer remotely: the customer photographs an ID document, then takes a live selfie. The server checks that the selfie is live and matches the ID photo, in one liveness-with-verification session.

sequenceDiagram participant App as Mobile app with liveness SDK participant Srv as Onboarding backend participant Face as Face API App->>Srv: Start onboarding, upload ID photo Srv->>Face: POST detectLivenessWithVerify-sessions with reference image Face-->>Srv: sessionId and authToken Srv-->>App: authToken only App->>Face: Liveness check with camera, using authToken Face-->>App: Done, no result on device App->>Srv: Session finished Srv->>Face: GET session result Face-->>Srv: livenessDecision realface, verify isIdentical true Srv->>Srv: Approve or route to manual review

The session token gives the device no access to results, so a tampered app can’t fake an approval. The decision is read only by the server.

TOKEN=$(az account get-access-token --resource https://cognitiveservices.azure.com --query accessToken -o tsv)

curl -s -X POST \
  "https://face-kyc-demo.cognitiveservices.azure.com/face/v1.2/detectLivenessWithVerify-sessions" \
  -H "Authorization: Bearer $TOKEN" \
  -F 'Parameters={"livenessOperationMode":"PassiveActive","deviceCorrelationId":"device-7f3a"};type=application/json' \
  -F "VerifyImage=@id-photo.jpg"

curl -s "https://face-kyc-demo.cognitiveservices.azure.com/face/v1.2/detectLivenessWithVerify-sessions/$SESSION_ID" \
  -H "Authorization: Bearer $TOKEN" | jq '.results.attempts[-1].result'

Source: Face liveness detection tutorial

from azure.ai.vision.face import FaceSessionClient
from azure.ai.vision.face.models import CreateLivenessWithVerifySessionContent, LivenessOperationMode

sessions = FaceSessionClient("https://face-kyc-demo.cognitiveservices.azure.com/", DefaultAzureCredential())
with open("id-photo.jpg", "rb") as ref:
    created = sessions.create_liveness_with_verify_session(
        CreateLivenessWithVerifySessionContent(liveness_operation_mode=LivenessOperationMode.PASSIVE_ACTIVE),
        verify_image=ref,
    )
print(created.session_id)  # send created.auth_token to the device

Source: Face liveness detection tutorial

Use cases

Scenario Feature Why this feature Example input → output

Scenario	Feature	Why this feature	Example input → output
KYC onboarding (finance)	Liveness with verification (Limited Access)	Remote identity proofing; spoof resistance.	ID photo + live selfie → `realface`, match confidence 0.93
Touchless building access	Identify against a large person group (Limited Access)	Employees enroll once; doors open on approach.	Camera frame → `personId` of an enrolled employee
Face blurring for privacy	Detect only (open to everyone)	Find faces to blur in published photos; no identification.	Event photo → bounding boxes to blur
Photo de-duplication	Find similar or group (Limited Access)	Clean up a member photo library.	Library → clusters of the same person
Bookshelf	Not used	No business need for biometrics; see the section above.	—

KYC onboarding (finance)

Liveness with verification (Limited Access)

Remote identity proofing; spoof resistance.

ID photo + live selfie → realface, match confidence 0.93

Touchless building access

Identify against a large person group (Limited Access)

Employees enroll once; doors open on approach.

Camera frame → personId of an enrolled employee

Face blurring for privacy

Detect only (open to everyone)

Find faces to blur in published photos; no identification.

Event photo → bounding boxes to blur

Photo de-duplication

Find similar or group (Limited Access)

Clean up a member photo library.

Library → clusters of the same person

Bookshelf

Not used

No business need for biometrics; see the section above.

—

When not to use it

When you only need to know whether an image contains people, Image Analysis people or a Foundry Model is enough, and avoids biometric processing.
For moderating profile pictures, use Content Safety.
Never use Face to infer emotion, gender or age; those attributes were retired.
See Choosing a Foundry Tool (or a Foundry Model) for alternatives.

Clean up

az cognitiveservices account delete --name face-kyc-demo --resource-group rg-kyc-demo
az group delete --name rg-kyc-demo --yes --no-wait

Source: Azure CLI reference — az cognitiveservices account delete

Delete person groups and face lists you no longer need; Face stores enrolled face templates until you delete them.

Networking, Security and Responsible AI covers Limited Access and transparency notes.
Vision: Image Analysis and OCR covers the rest of Azure Vision.
Responsible AI Practices covers impact assessment for sensitive uses.