2. Recognition Settings

2. Recognition Settings

Under Settings / Facial Recognition you will see the following settings:

Each of these individual settings is explained below.

No Touch Detection

With this setting turned on, NoahFace constantly scans for faces, and if it sees one, it will attempt to recognise it (ie: compare it with all its known faces). If you turn this setting off, you must first touch the screen before NoahFace will scan for faces.

The default is on. You should consider turning this setting off only if you are mounting NoahFace in a busy thoroughfare, and you don’t want it recognising faces that are passing by.

Detect Rotated Faces

With this setting turned on, NoahFace will recognise faces at up to a 30 degree angle. With this setting tuned off, NoahFace will only recognise faces that are directly facing the iPad screen.

The default is off, which minimises the chances of an incorrect recognition (aka ‘false positive’). You should consider turning this setting on if people cannot easily directly face the iPad screen due to its mounting location.

Detection Threshold

The detection threshold controls the process of finding faces in the video. If it is set to Very Low, if something in the image looks like it could be a face, it starts analysing it.

If it is set to Very High, it needs to be extremely clear that it looks like a face before it will start analysing it.

The default is Medium, which should work well in almost all cases.

You should consider changing this to Low if:

  • You have poor lighting conditions.
  • You have people with extremely dark skin or thick glasses that you are having trouble detecting.

You generally should not change this setting to Very Low – if you find you need this setting at Very Low to detect anyone, you should instead consider adjusting the camera brightness, physical shading, or installing front lighting.

You should consider changing this to High, Very High, or Maximum if:

  • You want to take every measure possible to minimise the chances of an incorrect recognition (aka ‘false positive’).

Detection Distance

This setting controls the distance at which the NoahFace App will recognise faces. As per the diagram below, this ranges from ‘Very Near’, which requires you to stand approximately 60cm from the iPad, to ‘Very Far’, which will detect faces as far away as 1.8 metres.

The default is Near, which will detect faces at approximately 90 cm.

You should consider changing this to Very Near if:

  • Your solution requires users to press buttons (eg: for Time and Attendance).
  • You are mounting your iPad in a busy thoroughfare, and you don’t want it recognising faces that are passing by.
  • You want to take every measure possible to minimise the chances of an incorrect recognition (aka ‘false positive’).

You should note that if you set this to Very Near and your iPad is mounted vertically, it may be difficult for extremely tall or extremely short people to get their faces in camera view.

You should consider changing this to Far or Very Far if:

  • You have a door at the end of a long approach, and you want to maximise the chances of the person being recognised before they reach the door.
  • You have people who are in wheelchairs.

Highlight Faces

With this setting turned on, NoahFace will draw a box around faces on the video screen.

Event Photos

An image of the user’s face is automatically clipped and stored with each event.

The options available are as follows:

Matching Threshold

Once a face is detected, NoahFace compares the face with all the faces of known individuals. This setting controls how confident the software needs to be that a particular face is a known individual before declaring it a match.

When most people hear about this threshold, their immediate response is that they want it set for absolute certainty. Unfortunately, this is not possible: there is a direct tradeoff between recognition rates and the chances of making a mistake. You should instead think of the Matching Threshold using a graph such as the following:

Note that this is no different from human recognition of faces. If you were to look at 1000 images containing faces of people you knew, and had to always declare who each image was, there would be some images where you could not be absolutely certain (eg: because the images were blurry or the person had their face turned to the side), and you could make mistakes. If you are allowed to pass on 20 images, the chances of you making a mistake are less than if you are only allowed to pass on 1 image. The only way to guarantee you never make a mistake is to pass on all images (which means you will achieve a 0% recognition rate).

The default setting of Medium is an extremely strong setting, and provides the right balance between excellent recognition and minimal chances of making a mistake in the vast majority of environments.

You should consider changing this to High, Very High, or Maximum if:

  • You want to take every measure possible to minimise the chances of an incorrect recognition (aka ‘false positive’), and you are happy to live with a lower recognition rate.

Biometric Learning

Biometric learning is the process of capturing and saving multiple sets of biometrics for the same person. These might be extracted at different distances, different angles, or in different lighting conditions, allowing NoahFace can achieve better recognition results.

NoahFace has three different biometric learning modes as follows:

  1. Registration. In this mode, biometric learning only happens during the registration process, when people manually identify themselves.
  2. Limited. In this mode, biometric learning happens automatically each time a person uses NoahFace for up to 10 days after they register.
  3. Continuous. In this mode, biometric learning happens automatically each time a person uses NoahFace. This is the default.

You should consider changing the Learning Mode to Registration or Limited if:

  • You want to take every measure possible to minimise the chances of an incorrect recognition (aka ‘false positive’), and you are happy to live with a lower recognition rate.

NoahFace also allows you to change the Learning Threshold, which is how certain it needs to be that a face belongs to a known person before additional biometrics are captured. The default is Medium.

You should consider changing the Learning Threshold to High, Very High, or Maximum if:

  • You want to take every measure possible to minimise the chances of an incorrect recognition (aka ‘false positive’), and you are happy to live with a lower recognition rate.

Self Correction

Once a match is determined, the NoahFace App will display the appropriate action screen – for example a Clock In/Out screen. If an individual has been incorrectly matched, they can press the ‘Not Me’ button so that the biometrics for the matched user are corrected. If you want to hide the ‘Not Me’ button, you can turn off this switch.