Azure Face Service: A Complete Guide to Face Detection, Liveness, and Recognition

The Azure Face service is a set of AI algorithms that detect, recognize, and analyze human faces in images. It supports a wide range of applications, from identification and touchless access control to automatic face blurring for privacy. Developers can access the service through client library SDKs or by calling the REST API directly, and can explore its capabilities quickly in Vision Studio.

Because facial recognition carries significant ethical risks, Face service access is limited based on eligibility and usage criteria and is available only to Microsoft-managed customers and partners. Applicants must complete the Face Recognition intake form. Microsoft has also retired or limited capabilities that infer emotional states or identity attributes, since misuse can lead to stereotyping, discrimination, or unfair denial of services. In June 2020, Microsoft announced it will not sell facial recognition technology to United States police departments until strong, human-rights-grounded regulation is enacted, and customers must acknowledge this restriction when creating a Face resource. Organizations processing biometric data are responsible for providing notice, obtaining consent, and deleting data as required by applicable law.

Common scenarios include verifying user identity against a trusted image from a government ID or enrollment photo, using liveness detection to confirm a real person is present, enabling touchless access control in airports, stadiums, offices, and schools to reduce card-sharing and security risks, and redacting or blurring faces in video to protect privacy. Working examples for customer check-in management and face photo tagging are available on GitHub.

Face detection is the foundational first step for all other scenarios. The Detect API returns the rectangle coordinates of faces along with a unique ID used in later identification or verification operations. Optionally, it can extract attributes such as head pose and glasses, which help applications ensure high-quality face data, for example by prompting users to remove sunglasses. Attributes like emotion and gender have been retired, and age, smile, facial hair, hair, and makeup are limited due to responsible AI concerns.

Liveness detection determines whether a face in a video stream is real or a spoof, defending against printed photos, recorded videos, and 2D or 3D masks. It is a critical component of biometric authentication systems that verify a physically present person. The solution achieved a 0% penetration rate in iBeta Level 1 and Level 2 Presentation Attack Detection tests, conformant to the ISO/IEC 30107-3 standard. Because liveness is a gated feature, customers must request access through the intake form before downloading the liveness SDKs for Android, iOS, or Web.

Identification performs one-to-many matching, comparing a detected face against a group of up to one million person objects, each holding up to 248 registered faces, and returns the matching person when found. Verification performs one-to-one matching to confirm whether two faces belong to the same person, supporting scenarios such as remote account opening or as a final check on identification results.

The Find Similar operation searches a set of candidate faces for those resembling a target face. The matchPerson mode filters results to the same individual using the Verify API, while matchFace mode returns similar faces regardless of identity. The Group operation divides an unordered set of unknown faces into disjoint groups based on similarity, returning a messyGroup array for faces without matches. Groups typically represent one person but may separate by factors such as expression.

Supported image formats are JPEG, PNG, GIF (first frame), and BMP, with a maximum file size of 6 MB. Faces must be between 36x36 and 4096x4096 pixels, and images larger than 1920x1080 require proportionally larger minimum face sizes. For recognition and verification, photos should be clear, sharp, and unaltered, with a single face in a neutral, front-facing pose, both eyes open, and mouth closed. Glasses, masks, hats, and head coverings should be avoided, facial jewelry must not hide the face, the background should be plain, and the face should fill at least 50% of the image and be well lit without shadows or red-eye.

Like all Foundry Tools resources, the Face service is subject to Microsoft's customer data policies. Users must understand these policies and comply with applicable biometric data protection requirements, which are detailed on the Foundry Tools page of the Microsoft Trust Center.

To begin building a face recognition application, follow a quickstart to code the basic components in your preferred language. Structured learning is also available through Microsoft Training modules for Face, and the Microsoft Certified: Azure AI Fundamentals certification covers foundational AI concepts for Azure.