Face identification

Biometric identification and face-based person search

Please comply with personal data protection laws! 

Usually, such a resource-intensive operation is performed on powerful video servers equipped with multiple NVIDIA graphics cards. Just try our small and affordable device for this task — and you will be truly surprised. 

The “Face Identification” package is provided as an add-on to the “ALL Neural Networks” package. Speclab offers two months of free testing. If you are not satisfied with the feature, you may simply choose not to pay for it, and the function will automatically deactivate. Try it for free!

Videoblazer includes two neural network functions: Face Recognition and Face Identification.

Differences

Recognition means detecting a face within video frames without calculating biometric metrics. It is simply an image of a face — without knowing whose face it is.

Identification performs biometric metric analysis in order to compare the face against stored biometric data of specific individuals. This allows identification by name.

Face Recognition is used for:

1. Highlighting events where faces appear — allowing operators to quickly track a person visually using facial images.

2. Capturing face frames that can later be entered into biometric analysis and used for face identification.

Face Identification allows you to:

• determine a person’s identity;

• trigger reactions when a specific individual appears;

• perform searches within archive databases.

The best identification results are achieved when a face is manually enrolled into the database in a well-lit environment, preferably against a uniform background.

The face should be held straight, without facial expressions, in front of a camera positioned at eye level. The “Capture” button may be used for an unlimited number of attempts.

Smiles and other facial expressions reduce the accuracy of biometric measurements.

When entering a name, you can also specify a group name in order to later assign algorithms or actions to the entire group. For example, for group “1”, the system can automatically unlock a door.

Immediately after this, the assigned name of the person appears on the image.

The entire system is now ready to operate with this identity. There are many possible use cases — for example, assigning a logical action:

If Camera 8 detects a face from the “ALL” group, … 

then activate Relay 1.

What is the error rate?

This is a common question. More technically advanced users often ask: What are the false positive and false negative rates?

The answer is: the question itself is not entirely correct. If you configure a high accuracy threshold and enable the “Frontal Faces Only” mode, there will be no errors at all.

However, in this case, the biometric metric will not work for all head positions and will only perform reliably in a well-lit environment. This option is better suited for access control systems.

If the accuracy threshold is reduced, the face will be detected from almost any angle and even in poor lighting conditions, but similar biometric matches from other people in the database may occasionally appear — especially if the database contains a large amount of data. As the number of enrolled faces increases, the probability of similarity also rises.

People wearing transparent glasses can also be identified, but it may be necessary to reduce the accuracy threshold.

Headwear does not affect the quality of identification in any way.

Faces can be added directly from the history of the “Attention!” panel.

Select a face detected by the face recognition system and click the “Add Face” button.

However, images captured in real-world conditions are usually obtained from less favorable angles compared to controlled posing conditions.

We should also answer in more detail the question: why is face recognition needed without identification?

Very often, video surveillance operates in dark environments with low resolution and poor face angles, where faces may be heavily rotated both vertically and horizontally, and may also be out of focus.

Under such conditions, the face identification system may fail to determine whose face it is, while a human operator may still recognize their employee even from behind.

Moreover, human intelligence can easily recognize artistic or symbolic images, such as silhouettes or outlines — something that is still beyond the capabilities of AI.

Very often, it is necessary to track people who entered a facility during the past week. Normally, this would require reviewing 24 hours × 7 days × 16 cameras = 2,688 hours of video footage. With face detection, however, the task can be completed in just a few minutes.