This guide is written on the assumption that AWS Custom Face Rekognition has been installed and configured correctly on your system. The installation guide can be found here.

For Custom Face Rekognition to work, the custom facial library must be built up. This is so that when a Custom Face Rekognition job is sent, it has a library to check against. To build the library, we will need to ingest stock images into the Curator system.

Here are some guidelines around the ingested images and the requirements for them to be used to build the library:

• Good results can be achieved from around a 200x200 pixel face in each shot.

• The image should not be very large (above 1920x1080), and can be much smaller.

• The image must not contain any other faces.

• The image should include all of the person’s head.

• The image should be facing forwards.

• If multiple images of the same person are used, they should all have the same library label.

Once you have ingested your stock images, you will need to create a collection, adding these images to the collection.

Each one of your images will need its relevant metadata added to it, on the image asset. 

My Example: I am ingesting a Stock image of Joe Root.

I need to add the metadata: FaceCollectionLabel: Joe Root

Once complete, send the created collection off to update the collection library using the AWS Rekognition plugin.

Using the AWS Rekognition Plugin, select the option to update the facial library.

This will send the Collection to the Mediastore: AWS-REKOGNITION-UPDATE-FACE-COLLECTION

Doing this will inform AWS that the images in the collection are to be associated with the metadata we created.

Now that we have updated our custom library with images, our AWS library should understand and will have learned who those faces are to be associated with.

We can test this by ingesting another image/video asset and sending the asset to our AWS Rekognition plugin, this time selecting Scan File for Custom Faces.

Once complete, we should see the result of the Rekognition job within the metadata on the asset.

The name of the face recognized should be displayed on the metadata: FaceSearchResults.