What makes a face creepy

Face recognition: clever or scary?

There are more than 25,000 digital photos on our hard drives - most of them show people. So far, the only way to find a desired face in this jumble of pixels has been through a manual search: by date, EXIM meta information, manually set key terms or simply by using your own memory. That is now changing: colleague computer will now help with the search. Smart face recognition algorithms, which the IT giants Apple and Google recently introduced and have now integrated into the latest versions of their photo management programs, make it possible.

Advances in the field of facial recognition have been made in recent years, especially by scientists who have dealt with anti-terrorist technologies. The idea: computers should automatically screen out attackers when they move through security checks at the airport, for example. However, it doesn't seem to work properly yet. In a test in Tampa, USA, airport employees who had previously voluntarily registered were only correctly identified in 53 percent of all cases. Civil rights groups also brought the risk of bad hits into the discussion: innocent people who suddenly find themselves there as terrorists and who might even be arrested because of their appearance. So it's no wonder that the technology is hardly discussed in public anymore.

The development is still not over. Many countries, including the United States, continue to work on the technology. For example, the specifications for passport photos have been overhauled there in order to make them easier to use for face recognition programs. The National Institute of Standards and Technology, which has been researching facial recognition software since 1994, conducted large tests in 2002 and 2006. Oregon and other US states began using facial recognition to screen out people trying to apply for a driver's license under a false name. All of this has resulted in the technology getting better and better over the past few years - much better.

For a face recognition system to work, a computer must first learn to distinguish between faces and other objects. Technically, this is easier to implement than the actual identification of individual persons. The approach was perfected shortly after the attacks of September 11, 2001. The result: systems for identifying faces are now being built into commercially available digital cameras and camcorders. These algorithms look for eyes, noses, and rounded areas of an image. A box is then defined on which the autofocus system can concentrate. Result: The image is always focused on faces, grandma’s eyes are never blurry anymore.

Face recognition therefore begins with identifying faces from the background in the first place. Then the face is rotated so that the eyes are on one plane and the image is brought to a standard size. Next, one of three technical approaches takes over. Each of them is based on its own patented ideas and is accordingly included in various software and hardware offers from different manufacturers. In one approach, the face is broken down into a mathematical pattern that can then be saved and searched, a second uses the entire face as a template and then searches for matches. Variant number three tries again to make a 3D model of the face in order to then look for geometric matches. In our experience with the software, Apple seems to focus on certain features on the face, while Google seems to match images. This is not clear, however, because none of the companies discloses which algorithms are actually used.

Read comments (2) Go to homepage