Biometric Face Recognition, the Perfect Task for Artificial Intelligence

Senator Torsten Klawunder

Senator Torsten Klawunder

Author Biography

CEO - DiVis Hygiene GmbH | Senator of the European Senate of Economy and Technology

 

Over the years, artificial intelligence (AI) is a term which has become more and more popular in views of innovative software solutions for different areas of technology.

What is behind it and which innovations already exist on this basis? As the name suggests, it is a self-learning computer software (artificial intelligence) based on neural networks that are modeled similar to structures utilised in the human brain. The idea is not new, but has only been effectively implemented in recent years thanks to the further increase in computing power, and affordable large storage capacities of today’s computer systems.

Because learning for yourself means, of course, that a system has to be presented with a lot of data with which a model is trained, and a later decision-making process has to go through an immense number of mathematical arithmetic operations, in order to react in real time within milliseconds. Autonomous driving is a classic example of one of the most complex challenges that are currently being tried to be solved with artificial intelligence. For a long time, this technology was reserved for research laboratories, universities and a few specialist companies. However, since software tools were created with which it is relatively easy to train your own models, new areas of application are constantly being developed and existing solutions are being perfected. Let’s just think of Amazon Alexa and compare that with speech recognition systems as we have known them up to now.

This is just a small background to have a little idea of the possibilities that intelligent AI-based computer solutions will bring with them in the future.

As a more specific example of this, I would like to look at person recognition based on biometric facial features.

More than 20 years ago, computers began to be used to identify people on the basis of photos of their faces. Even if the results were still relatively flawed, the computer-aided person search in forensics made work considerably easier.

Right from the start, search algorithms were set up using neural networks and have been optimized in a self-learning manner using training algorithms. In principle, this was already based on artificial intelligence, and the basic idea has remained more or less the same as it is today.

Using a virtual grid pattern that is placed over the face, various striking points in terms of shape, size, angle, etc. are optically measured with respect to one another and assigned to a person as a mathematical model. It is preferable to use features that change little. Features such as hairstyle, hair color, beard are not suitable for biometric identification, as these features change frequently. Even areas that change due to changes in feelings (e.g. when laughing) are still partially excluded as facial features.

Till today, people are asked to look seriously when taking photos for a passport.

Although the algorithms for face recognition have been not so different 10 years ago, as they are today, there are worlds between the current very high quality of face recognition and the high error rates 10 years ago.

Today we are talking about a risk of confusion with <0.001% with a detection rate of >99.9%. This, however, on the condition that a facial photo of the person is available, which is sufficiently exposed and the person is looking straight into the camera.

With the help of powerful computers and optimized databases, even searching for people in a database with 30,000,000 faces within a second is no longer a utopia.

How can it be explained that the quality, handling, speed and above all security of face recognition has made such immense progress in a relatively short period of time that the computer is now far superior to humans in this area.

An extreme example is the fact that a good face algorithm will even accept faces that are covered by a medical face mask, and identify such people in 98%.

A very decisive factor for this development is, of course, the constant improvement in the computing power and storage capacity of computers, combined with standard software for training models based on artificial intelligence.

Models based on artificial intelligence are no longer reserved for universities and research centers, but have also found their way into many software development departments.

But the most crucial point with artificial intelligence is the amount of material with which an AI model can be trained.

In terms of face recognition, it was helpful that, over the past few years, more and more face recognition systems have been installed around the world, in which the users have released their image recordings for AI training and thus today, millions of face recordings from all ethnic groups are available for training AI models. For this reason, the biometric face recognition is no longer just an aid in criminology and passport control, but is increasingly used in everyday situations.

Examples of facial recognition applications outside of forensics, criminalistics and border control:

  • Access control systems for stadiums, events, amusement parks, zoos, cruise ships, airports, security areas, companies, etc;
  • PIN code based unlocking of computers, smart phones, etc;
  • Payments, ATMs; and
  • Access to all areas where you previously required e.g. an entrance ticket, boarding pass, ID, key, pin code or others.

The great advantage of face recognition compared to fingerprint, vein scanner, iris recognition, speech recognition and pin code entry is the fact that a face photo can be taken without special action, contactless, and even from a distance – practically on the move. This works very fast and is convenient for the user. However, an automatically working image recording system without human assistance needs an intelligent software which takes over the monitoring and control of the image recordings.

Artificial intelligence is also an option here:

  • Search for faces in an image.
  • Check whether the person is in a predefined area and has a defined position.
  • Check whether the facial image meets certain criteria and is of sufficient quality. (e.g. size of the face image, resolution of the image recording, exposure, focus, head position, etc.)
  • Is it a real face at all, or an attempt is being made to fool the system. (e.g. with a face mask, or a face photo which is held in front of the face, or similar)

As already mentioned, facial recognition can be used in many areas to make processes easier for users, to minimize waiting times, to make access more secure, etc.

A major obstacle, however, is often objections to data protection and concerns of the user regarding misuse of their own image, which is ultimately handed over for this convenience.
Yes, I understand reservations and I myself reject projects in which the user cannot refuse to use his or her image. However, this does not apply to the verification of identity documents such as passports, driving licenses, etc. Here, no face photo or template is stored, and only for a short moment, the verification between the photo on the document and the live image of the person is processed.

Nevertheless, I would like to go into some processes in order to minimize general fears of abuse.

In general, no special image is required for face recognition. The old title “biometric image” actually fueled fears, which is complete nonsense, since it is actually nothing more than a ‘selfie’. Only in the case of official documents is it precisely defined in which format the photo has to be, that it is for instance, taken from the front or that one should not smile.

The captured photo is not needed for the face recognition itself and there is no need to store it for biometric verification or identification. For the face recognition itself, only a so-called template is generated out of this face image. The template only contains the mathematical relationships between certain facial features. This template cannot be converted back into the face image. The face image itself can be deleted after the template generation.

The template itself can be crypted and in a case of a personalized entrance card, the template can be printed in a QR code on the entrance card.

With all this, a face recognition system can be realized in compliance to data protection laws.

A modern face recognition solution, is perfect to combine security with user comfort and quick check in.

Female Empowerment in the Digital Age

Dr. Laura Bechthold is a social scientist and innovation professional from Munich. As a postdoctoral researcher at the Friedrichshafen Institute for Family Entrepreneurship at Zeppelin University, she works on questions regarding responsibility and decision paradigms of family entrepreneurs. As the Director of Science Services at Philoneos GmbH, she supports family fi rms in establishing organizational structures for innovation. Laura holds a BA in Business Administration (Zeppelin University), a Master of Business Research (LMU Munich) and an MSc in Sustainability Science and Policy (Maastricht University). Her PhD research focused on unconscious biases in female entrepreneurship. Her fi eld experimental study on female entrepreneurial role models was awarded twice at international conferences. Laura’s passion lies in building bridges between science and practice to foster an open dialogue and co-create solutions for an inclusive, sustainable and prospering society. Therefore, she contributes to EUTECH by writing about entrepreneurial challenges and opportunities for contributing to the Sustainable Development Goals.

Read More »