Azure AI Face Service

Quick Answer: Azure AI Face provides face detection (locate faces and attributes), verification (1:1 — are these the same person?), identification (1:N — who is this person?), and liveness detection (is this a live person vs. a photo/video?). Identification and verification require Microsoft approval for access.

Face Detection

Face detection locates faces in an image and returns face rectangles, landmarks, and optional attributes:

from azure.ai.vision.face import FaceClient
from azure.ai.vision.face.models import (
    FaceDetectionModel,
    FaceRecognitionModel,
    FaceAttributeTypeDetection03,
    FaceAttributeTypeRecognition04
)
from azure.core.credentials import AzureKeyCredential

client = FaceClient(
    endpoint="https://my-face.cognitiveservices.azure.com/",
    credential=AzureKeyCredential("<your-key>")
)

# Detect faces with attributes
with open("photo.jpg", "rb") as f:
    detected_faces = client.detect(
        image_content=f.read(),
        detection_model=FaceDetectionModel.DETECTION_03,
        recognition_model=FaceRecognitionModel.RECOGNITION_04,
        return_face_id=True,
        return_face_landmarks=True,
        return_face_attributes=[
            FaceAttributeTypeDetection03.HEAD_POSE,
            FaceAttributeTypeDetection03.BLUR,
            FaceAttributeTypeDetection03.MASK
        ]
    )

for face in detected_faces:
    rect = face.face_rectangle
    print(f"Face at: ({rect.left}, {rect.top}), "
          f"size: {rect.width}x{rect.height}")
    if face.face_attributes:
        print(f"  Head pose: {face.face_attributes.head_pose}")

Detection Models

Recognition Models

On the Exam: Always pair detection_03 with recognition_04 for the best results. Questions may ask which model combination provides the highest accuracy.

Face Verification (1:1)

Face verification compares two faces to determine if they belong to the same person:

# Detect faces in two images
face1 = client.detect(image_content=image1_bytes,
                       return_face_id=True)[0]
face2 = client.detect(image_content=image2_bytes,
                       return_face_id=True)[0]

# Verify if they are the same person
result = client.verify_face_to_face(
    face_id1=face1.face_id,
    face_id2=face2.face_id
)

print(f"Same person: {result.is_identical}")
print(f"Confidence: {result.confidence:.2f}")

Use cases: Identity verification (ID card photo vs. selfie), access control, KYC (Know Your Customer) processes.

Face Identification (1:N)

Face identification matches an unknown face against a group of known faces:

Step 1: Create a PersonGroup

# Create a person group
client.create_person_group(
    person_group_id="employees",
    name="Company Employees",
    recognition_model=FaceRecognitionModel.RECOGNITION_04
)

Step 2: Add People and Their Face Images

# Add a person to the group
person = client.create_person_group_person(
    person_group_id="employees",
    name="Jane Doe"
)

# Add face images for the person (multiple images improve accuracy)
with open("jane_photo1.jpg", "rb") as f:
    client.add_person_group_person_face(
        person_group_id="employees",
        person_id=person.person_id,
        image_content=f.read()
    )

Step 3: Train the PersonGroup

# Train the person group (required after adding/removing faces)
client.train_person_group(person_group_id="employees")

# Check training status
status = client.get_person_group_training_status(
    person_group_id="employees"
)
print(f"Training status: {status.status}")

Step 4: Identify Unknown Faces

# Detect a face in a new image
unknown_face = client.detect(
    image_content=unknown_photo_bytes,
    return_face_id=True
)[0]

# Identify against the person group
results = client.identify(
    face_ids=[unknown_face.face_id],
    person_group_id="employees"
)

for result in results:
    if result.candidates:
        best_match = result.candidates[0]
        print(f"Match: person {best_match.person_id}")
        print(f"Confidence: {best_match.confidence:.2f}")
    else:
        print("No match found")

Liveness Detection

Liveness detection verifies that the face input is from a live person (not a photo, video, or deepfake):

On the Exam: Liveness detection is a Responsible AI safeguard. Questions may describe a scenario where an attacker holds up a photo to bypass face verification — the answer is to add liveness detection.

Responsible Use Restrictions

Microsoft restricts access to certain Face API features:

Important: Microsoft retired emotion recognition and demographic attributes (age, gender) from the Face API in 2023 as part of their Responsible AI commitments. The AI-102 exam may test whether you know these features are no longer available.