“Face recognition” or “facial recognition” terms are usually used to highlight biometric algorithms and technologies for identifying an individual on a live or still image in comparison with another image of an individual based on human face analysis. Most facial recognition systems use 2D projections of 3D scenes to perform face analysis.
Facial recognition algorithms can use different mathematical approaches, and mostly they use at least 2 stages starting from detection of a face region on a given image to the creation of a faceprint byte array to be used in further analysis.
Often, facial recognition systems use a faceprint term, which is a digitally recorded representation of a person’s face. By other words this is a key fingerprint for person’s face.
Each human face according to 3D analysis has distinguishable landmarks known as nodal points. Human face has about 80 nodal points.
Face’s nodal points are endpoints which help measuring certain variables of a person’s face, including the width or length of the nose, space between the eyes etc. The full set of such measures constitutes the faceprint. In our facial recognition algorithms we often use the “face features” term, which is a 512 byte array representing a synonym for faceprint term. A “key fingerprint” term for face’s measures is a synonym for faceprint too.
The facial identification algorithm is basically a comparison of two faceprints. Faceprints always include small deviations and cannot be calculated with 100% of certainty, thus the faceprints comparison result is always a probability-based value.
Nevertheless, modern facial recognition software reaches approximately 99.5% accuracy, which is reliable result.
Face Recognition Terminals
Human Face Nodal Points
In general, facial recognition algorithms use different mathematical methods but due to fast recent development of microprocessors and microcontrollers, Artificial Intelligence mathematical methods, i.e. ai face recognition algorithms, become widely in use.
Commonly, AI methods consist of 2 phases as follows:
- ai face recognition algorithm deep learning;
- ai face recognition algorithm production use;
Several common steps of face recognition algorithms are briefly described below.
1. Detection of face region
First, the program detects a person’s face region, whether they are alone or in a crowd. It is easier to detect a face when the person is looking directly into the camera, but it is also possible to detect a face in situations where a person does not look in the camera.
2. Nodal points discovery
Second, the system starts the analysis.
Algorithms analyze the nodes between individual’s face 80 nodal points: distance between the eyes, shape of cheekbones, of nose and lips.
3. Faceprint calculations
Third, the system converts the results of face analysis into a mathematical formula. Face features become a numeric code, called a faceprint. Similar to the unique structure of the fingerprint, each person has their own faceprint.
A faceprint is a quite accurate representation of real person’s face features. However, some deviations (unimportant most of the time) are always produced, so we cannot speak about a 100% match.
4. Faces comparison
Next, the created faceprint is compared with each faceprint in database.
Program determines whether faceprint data matches the data found in database with a certain accuracy degree and gives the result of identification – for example, by providing user information.
Nowadays facial recognition have often over 99% accuracy.
It is becomes more and more often, user identification software along with access control systems use face recognition or face recognition technologies in a way to obtain a simple answer: can a person from a live image can be identified with anyone who has access granted.
Nowadays, facial recognition software become widely in use in different products as well as a part of turnkey solutions and services. For example, today’s video surveillance cameras can use embedded face recognition algorithms. Global market offers small and affordable development cards with ai face recognition technology at price tag below 50.- CHF / USD.
The use of facial recognition algorithm in network services or websites won’t require any camera at all as it recognizes faces on given pictures as a service – Facebook is a good example of such a use for friends tagging.
Modern facial recognition technology includes anti-spoofing, so it becomes more and more difficult to cheat the face recognition machine. If a user tries to pretend being another person by wearing a mask or putting a printed picture on their face, the program will detect the problem. Detection and recognition speed is very fast: usually identification takes few milliseconds. Moreover, facial recognition starts working even at low light conditions and at several meters distance to the face. Modern algorithms are also able to detect more than one face at a time.
Facial recognition software has a wide application field, from tagging friends and smartphone Face IDs to forensics. They are even used in medicine – for instance, to detect genetic disorders.
Facial recognition system offers several advantages such as high accuracy and absence of physical contact: therefore we talk about a contactless technology, so, a user doesn’t need to touch the device, unlike the use of other biometric technologies such as fingerprint, or physical input an access code on keypad.
These benefits have been highlighted and approved during Covid-19 pandemic, when everyone needs to reduce the number of physical contacts.
Today, there are many open source algorithms and systems providing face recognition algorithms for free and thus spreading them out in the wide use. In the same time, there are many providers who evolve open source algorithms or develop proprietary algorithms with deep learning training results and offer their results in a form of face recognition library (face recognition SDK). The use of such libraries dramatically reduces the integration cost in final products and services.
ThermoVSN access control system is developed in a flexible and generic way to enable any integration of near any possible or existing face recognition library from any provider available on global market.
Face recognition technology is a form of biometric recognition technology along with fingerprint recognition, iris recognition, hand geometry recognition, voice recognition and more.
Nowadays biometric identification and authorization methods become more and more in the use. However, the wide use of biometric data rises big questions about data privacy and protection.
Basically, personal data can be stored on a device or a server, like, for instance, your organization’s server. In this case, your company is responsible for data storage, processing, non-disclosure, protection etc.
In Europe, basic principles of data protection are described in General Data Protection Regulation (GDPR):
- Lawfulness, fairness and transparency
- Purpose limitation
- Data minimization
- Storage limitation
- Integrity and confidentiality (security)
If you are in the United States, check the California Consumer Privacy Act regulations.
ThermoVSN access control devices are an example of what a turnkey solution with face recognition is. ThermoVSN Face X terminals have an embedded binocular camera, combining RGB and IR video streams. Face recognition software works along with both video streams. Our units use multifactor access control authorization and facial recognition is one of such factors. The embedded application developed by Swiss Biometrix is a flexible and adapted to customer needs software, specially designed for access management.
- 99.7% recognition accuracy
- 0.2 s recognition speed
- Up to 2 m recognition distance
- ≥100dB wide dynamic range – fits for environments with complex light
- Works in low light
- Can support multiple recognition algorithms
- Face database can contain up to 100’000 pictures
- GDPR compliant data protection
- Decentralized data storage (option)
- Face recognition fits for multi-factor authentication
(can be combined with other authentication modes)
- Face recognition algorithms prevent prohibited access
In ThermoVSN Face X terminals, even if the face recognition doesn’t work as authorization factor, it keeps operating to ensure there is a live person (face) inside of analyzed video frame.
This means that if for some reason the device had to detect credentials such as QR code or a badge with no person in front of the camera, the system would not grant the access. Thus, we also use face recognition as fraud prevention.