3. Liveness Detection

The aim of liveness detection is to differentiate real people from photographs and videos. It is used to avoid fraud / spoofing attempts. NoahFace has three liveness testing algorithms built-in, which can be used independently or together by turning on the appropriate switches on the Settings/Facial Recognition page. Each of these algorithms is explained below:

Movement Analysis

This algorithm attempts to analyse the movement in the image (frame to frame) to determine if the face is a real face or not. The edges of the face are particularly important to this algorithm, which makes it sensitive to the background. It works exceptionally well if the camera is facing a blank white wall. However, it can be problematic if it is facing a complex background, such as a window, shelving, or even a picture on the wall.

‍

Constant Background

This algorithm compares the background surrounding a face with the background of the video feed when there are no faces present. If the it differs significantly, then it assumes that someone is attempting to fool the recognition process with a photograph. This test is not at all sensitive to the complexity of the background, as long as it is static. However, this test will ONLY work if:

• You have the iPad mounted on the wall so it cannot move.
• You have consistent lighting and you are using a fixed shutter speed.
• You do NOT have people, cars, or anything else constantly moving in the background.

‍

Depth Analysis

This algorithm examines the depth of a face at various points to ensure it is three-dimensional. This algorithm relies on the availability of a TrueDepth camera, which is currently only available on an iPad Pro 2018 model or later. The T rueDepth camera can only accurately determine depth at up to 1.2 meters, so faces further away than this will not be recognised regardless of the Detection Distance and Zoom settings.

‍

IMPORTANT - If detecting fraud is critical to your application, we strongly recommend considering purchasing an iPad Pro and using the Depth Analysis liveness test.

‍

Liveness Testing Operation

When NoahFace believes it has detected fraud (ie: not a real person), the recognition process will pause for the configured Lock Time. If this happens you will see that the Enter button on the Welcome screen remains disabled, and the screen will flash red if you have Visual Feedback enabled.

‍

Setting the Lock Time too low can compromise the liveness testing, so we recommend not setting it lower than 3 seconds, which is the default.

‍

Preparing your Environment

If you plan to use Liveness Testing, please consider:

1. The liveness testing algorithms will NOT work reliably if your iPad is not mounted.
2. If you are using the Movement Analysis or Constant Background algorithms, the location you mount your iPad is very important. You will get the best results if your iPad camera is facing a blank wall, and you may get poor results if your iPad camera is facing a complex background (eg: a doorway, a bookcase, or a picture on the wall).
3. When you start the NoahFace App, you MUST wait for the message “Preparing liveness detection...” to disappear before it is used. If you do not wait for this message to disappear, then the liveness testing algorithms will NOT work reliably.
4. If detecting fraud is critical to your application, we strongly recommend considering purchasing an iPad Pro and using the Depth Analysis liveness test. This is the strongest form of liveness testing – that is, it will detect the most cases.

‍