Use the PersonDirectory structure

To perform face recognition operations such as Identify and Find Similar, Face API customers need to create an assorted list of Person objects. The new PersonDirectory is a data structure that contains unique IDs, optional name strings, and optional user metadata strings for each Person identity added to the directory.

Currently, the Face API offers the LargePersonGroup structure, which has similar functionality but is limited to 1 million identities. The PersonDirectory structure can scale up to 75 million identities.

Another major difference between PersonDirectory and previous data structures is that you'll no longer need to make any Train calls after adding faces to a Person object—the update process happens automatically.

Prerequisites

  • Azure subscription - Create one for free.
  • Once you have your Azure subscription, create a Face resource in the Azure portal to get your key and endpoint. After it deploys, click Go to resource.
    • You'll need the key and endpoint from the resource you create to connect your application to the Face API. You'll paste your key and endpoint into the code below.
    • You can use the free pricing tier (F0) to try the service, and upgrade later to a paid tier for production.

Add Persons to the PersonDirectory

Persons are the base enrollment units in the PersonDirectory. Once you add a Person to the directory, you can add up to 248 face images to that Person, per recognition model. Then you can identify faces against them using varying scopes.

Create the Person

To create a Person, you need to call the CreatePerson API and provide a name or userData property value.

using Newtonsoft.Json;
using System;
using System.Collections.Generic;
using System.Linq;
using System.Net.Http;
using System.Net.Http.Headers;
using System.Text;
using System.Threading.Tasks;

var client = new HttpClient();

// Request headers
client.DefaultRequestHeaders.Add("Ocp-Apim-Subscription-Key", "{subscription key}");

var addPersonUri = "https:// {endpoint}/face/v1.0-preview/persons";

HttpResponseMessage response;

// Request body
var body = new Dictionary<string, object>();
body.Add("name", "Example Person");
body.Add("userData", "User defined data");
byte[] byteData = Encoding.UTF8.GetBytes(JsonConvert.SerializeObject(body));

using (var content = new ByteArrayContent(byteData))
{
    content.Headers.ContentType = new MediaTypeHeaderValue("application/json");
    response = await client.PostAsync(addPersonUri, content); 
}

The CreatePerson call will return a generated ID for the Person and an operation location. The Person data will be processed asynchronously, so you use the operation location to fetch the results.

Wait for asynchronous operation completion

You'll need to query the async operation status using the returned operation location string to check the progress.

First, you should define a data model like the following to handle the status response.

[Serializable]
public class AsyncStatus
{
    [DataMember(Name = "status")]
    public string Status { get; set; }

    [DataMember(Name = "createdTime")]
    public DateTime CreatedTime { get; set; }

    [DataMember(Name = "lastActionTime")]
    public DateTime? LastActionTime { get; set; }

    [DataMember(Name = "finishedTime", EmitDefaultValue = false)]
    public DateTime? FinishedTime { get; set; }

    [DataMember(Name = "resourceLocation", EmitDefaultValue = false)]
    public string ResourceLocation { get; set; }

    [DataMember(Name = "message", EmitDefaultValue = false)]
    public string Message { get; set; }
}

Using the HttpResponseMessage from above, you can then poll the URL and wait for results.

string operationLocation = response.Headers.GetValues("Operation-Location").FirstOrDefault();

Stopwatch s = Stopwatch.StartNew();
string status = "notstarted";
do
{
    if (status == "succeeded")
    {
        await Task.Delay(500);
    }

    var operationResponseMessage = await client.GetAsync(operationLocation);

    var asyncOperationObj = JsonConvert.DeserializeObject<AsyncStatus>(await operationResponseMessage.Content.ReadAsStringAsync());
    status = asyncOperationObj.Status;

} while ((status == "running" || status == "notstarted") && s.Elapsed < TimeSpan.FromSeconds(30));

Once the status returns as "succeeded", the Person object is considered added to the directory.

Note

The asynchronous operation from the Create Person call does not have to show "succeeded" status before faces can be added to it, but it does need to be completed before the Person can be added to a DynamicPersonGroup (see below Create and update a DynamicPersonGroup) or compared during an Identify call. Verify calls will work immediately after faces are successfully added to the Person.

Add faces to Persons

Once you have the Person ID from the Create Person call, you can add up to 248 face images to a Person per recognition model. Specify the recognition model (and optionally the detection model) to use in the call, as data under each recognition model will be processed separately inside the PersonDirectory.

The currently supported recognition models are:

  • Recognition_02
  • Recognition_03
  • Recognition_04

Additionally, if the image contains multiple faces, you'll need to specify the rectangle bounding box for the face that is the intended target. The following code adds faces to a Person object.

var client = new HttpClient();

// Request headers
client.DefaultRequestHeaders.Add("Ocp-Apim-Subscription-Key", "{subscription key}");

// Optional query strings for more fine grained face control
var queryString = "userData={userDefinedData}&targetFace={left,top,width,height}&detectionModel={detectionModel}";
var uri = "https://{endpoint}/face/v1.0-preview/persons/{personId}/recognitionModels/{recognitionModel}/persistedFaces?" + queryString;

HttpResponseMessage response;

// Request body
var body = new Dictionary<string, object>();
body.Add("url", "{image url}");
byte[] byteData = Encoding.UTF8.GetBytes(JsonConvert.SerializeObject(body));

using (var content = new ByteArrayContent(byteData))
{
    content.Headers.ContentType = new MediaTypeHeaderValue("application/json");
    response = await client.PostAsync(uri, content);
}

After the Add Faces call, the face data will be processed asynchronously, and you'll need to wait for the success of the operation in the same manner as before.

When the operation for the face addition finishes, the data will be ready for in Identify calls.

Create and update a DynamicPersonGroup

DynamicPersonGroups are collections of references to Person objects within a PersonDirectory; they're used to create subsets of the directory. A common use is when you want to get fewer false positives and increased accuracy in an Identify operation by limiting the scope to just the Person objects you expect to match. Practical use cases include directories for specific building access among a larger campus or organization. The organization directory may contain 5 million individuals, but you only need to search a specific 800 people for a particular building, so you would create a DynamicPersonGroup containing those specific individuals.

If you've used a PersonGroup before, take note of two major differences:

  • Each Person inside a DynamicPersonGroup is a reference to the actual Person in the PersonDirectory, meaning that it's not necessary to recreate a Person in each group.
  • As mentioned in previous sections, there is no need to make Train calls, as the face data is processed at the Directory level automatically.

Create the group

To create a DynamicPersonGroup, you need to provide a group ID with alphanumeric or dash characters. This ID will function as the unique identifier for all usage purposes of the group.

There are two ways to initialize a group collection. You can create an empty group initially, and populate it later:

var client = new HttpClient();

// Request headers
client.DefaultRequestHeaders.Add("Ocp-Apim-Subscription-Key", "{subscription key}");

var uri = "https://{endpoint}/face/v1.0-preview/dynamicpersongroups/{dynamicPersonGroupId}";

HttpResponseMessage response;

// Request body
var body = new Dictionary<string, object>();
body.Add("name", "Example DynamicPersonGroup");
body.Add("userData", "User defined data");
byte[] byteData = Encoding.UTF8.GetBytes(JsonConvert.SerializeObject(body));

using (var content = new ByteArrayContent(byteData))
{
    content.Headers.ContentType = new MediaTypeHeaderValue("application/json");
    response = await client.PutAsync(uri, content);
}

This process is immediate and there is no need to wait for any asynchronous operations to succeed.

Alternatively, you can create it with a set of Person IDs to contain those references from the beginning by providing the set in the AddPersonIds argument:

var client = new HttpClient();

// Request headers
client.DefaultRequestHeaders.Add("Ocp-Apim-Subscription-Key", "{subscription key}");

var uri = "https://{endpoint}/face/v1.0-preview/dynamicpersongroups/{dynamicPersonGroupId}";

HttpResponseMessage response;

// Request body
var body = new Dictionary<string, object>();
body.Add("name", "Example DynamicPersonGroup");
body.Add("userData", "User defined data");
body.Add("addPersonIds", new List<string>{"{guid1}", "{guid2}", …});
byte[] byteData = Encoding.UTF8.GetBytes(JsonConvert.SerializeObject(body));

using (var content = new ByteArrayContent(byteData))
{
    content.Headers.ContentType = new MediaTypeHeaderValue("application/json");
    response = await client.PutAsync(uri, content);

    // Async operation location to query the completion status from
    var operationLocation = response.Headers.Get("Operation-Location");
}

Note

As soon as the call returns, the created DynamicPersonGroup will be ready to use in an Identify call, with any Person references provided in the process. The completion status of the returned operation ID, on the other hand, indicates the update status of the person-to-group relationship.

Update the DynamicPersonGroup

After the initial creation, you can add and remove Person references from the DynamicPersonGroup with the Update Dynamic Person Group API. To add Person objects to the group, list the Person IDs in the addPersonsIds argument. To remove Person objects, list them in the removePersonIds argument. Both adding and removing can be performed in a single call:

var client = new HttpClient();

// Request headers
client.DefaultRequestHeaders.Add("Ocp-Apim-Subscription-Key", "{subscription key}");

var uri = "https://{endpoint}/face/v1.0-preview/dynamicpersongroups/{dynamicPersonGroupId}";

HttpResponseMessage response;

// Request body
var body = new Dictionary<string, object>();
body.Add("name", "Example Dynamic Person Group updated");
body.Add("userData", "User defined data updated");
body.Add("addPersonIds", new List<string>{"{guid1}", "{guid2}", …});
body.Add("removePersonIds", new List<string>{"{guid1}", "{guid2}", …});
byte[] byteData = Encoding.UTF8.GetBytes(JsonConvert.SerializeObject(body));

using (var content = new ByteArrayContent(byteData))
{
    content.Headers.ContentType = new MediaTypeHeaderValue("application/json");
    response = await client.PatchAsync(uri, content);

    // Async operation location to query the completion status from
    var operationLocation = response.Headers.Get("Operation-Location");
}

Once the call returns, the updates to the collection will be reflected when the group is queried. As with the creation API, the returned operation indicates the update status of person-to-group relationship for any Person that's involved in the update. You don't need to wait for the completion of the operation before making further Update calls to the group.

Identify faces in a PersonDirectory

The most common way to use face data in a PersonDirectory is to compare the enrolled Person objects against a given face and identify the most likely candidate it belongs to. Multiple faces can be provided in the request, and each will receive its own set of comparison results in the response.

In PersonDirectory, there are three types of scopes each face can be identified against:

Scenario 1: Identify against a DynamicPersonGroup

Specifying the dynamicPersonGroupId property in the request compares the face against every Person referenced in the group. Only a single DynamicPersonGroup can be identified against in a call.

var client = new HttpClient();

// Request headers
client.DefaultRequestHeaders.Add("Ocp-Apim-Subscription-Key", "{subscription key}");

// Optional query strings for more fine grained face control
var uri = "https://{endpoint}/face/v1.0-preview/identify";

HttpResponseMessage response;

// Request body
var body = new Dictionary<string, object>();
body.Add("faceIds", new List<string>{"{guid1}", "{guid2}", …});
body.Add("dynamicPersonGroupId", "{dynamicPersonGroupIdToIdentifyIn}");
byte[] byteData = Encoding.UTF8.GetBytes(JsonConvert.SerializeObject(body));

using (var content = new ByteArrayContent(byteData))
{
    content.Headers.ContentType = new MediaTypeHeaderValue("application/json");
    response = await client.PostAsync(uri, content);
}

Scenario 2: Identify against a specific list of persons

You can also specify a list of Person IDs in the personIds property to compare the face against each of them.

var client = new HttpClient();

// Request headers
client.DefaultRequestHeaders.Add("Ocp-Apim-Subscription-Key", "{subscription key}");
            
var uri = "https://{endpoint}/face/v1.0-preview/identify";

HttpResponseMessage response;

// Request body
var body = new Dictionary<string, object>();
body.Add("faceIds", new List<string>{"{guid1}", "{guid2}", …});
body.Add("personIds", new List<string>{"{guid1}", "{guid2}", …});
byte[] byteData = Encoding.UTF8.GetBytes(JsonConvert.SerializeObject(body));

using (var content = new ByteArrayContent(byteData))
{
    content.Headers.ContentType = new MediaTypeHeaderValue("application/json");
    response = await client.PostAsync(uri, content);
}

Scenario 3: Identify against the entire PersonDirectory

Providing a single asterisk in the personIds property in the request compares the face against every single Person enrolled in the PersonDirectory.

var client = new HttpClient();

// Request headers
client.DefaultRequestHeaders.Add("Ocp-Apim-Subscription-Key", "{subscription key}");
            
var uri = "https://{endpoint}/face/v1.0-preview/identify";

HttpResponseMessage response;

// Request body
var body = new Dictionary<string, object>();
body.Add("faceIds", new List<string>{"{guid1}", "{guid2}", …});
body.Add("personIds", new List<string>{"*"});
byte[] byteData = Encoding.UTF8.GetBytes(JsonConvert.SerializeObject(body));

using (var content = new ByteArrayContent(byteData))
{
    content.Headers.ContentType = new MediaTypeHeaderValue("application/json");
    response = await client.PostAsync(uri, content);
}

For all three scenarios, the identification only compares the incoming face against faces whose AddPersonFace call has returned with a "succeeded" response.

Verify faces against persons in the PersonDirectory

With a face ID returned from a detection call, you can verify if the face belongs to a specific Person enrolled inside the PersonDirectory. Specify the Person using the personId property.

var client = new HttpClient();

// Request headers
client.DefaultRequestHeaders.Add("Ocp-Apim-Subscription-Key", "{subscription key}");

var uri = "https://{endpoint}/face/v1.0-preview/verify";

HttpResponseMessage response;

// Request body
var body = new Dictionary<string, object>();
body.Add("faceId", "{guid1}");
body.Add("personId", "{guid1}");
var jsSerializer = new JavaScriptSerializer();
byte[] byteData = Encoding.UTF8.GetBytes(jsSerializer.Serialize(body));

using (var content = new ByteArrayContent(byteData))
{
    content.Headers.ContentType = new MediaTypeHeaderValue("application/json");
    response = await client.PostAsync(uri, content);
}

The response will contain a Boolean value indicating whether the service considers the new face to belong to the same Person, and a confidence score for the prediction.

Next steps

In this guide, you learned how to use the PersonDirectory structure to store face and person data for your Face app. Next, learn the best practices for adding your users' face data.