The needle in the haystack
Everyone knows how difficult it is to find a certain face in a crowd of people. New software is now showing good results in recognizing certain faces in the so-called "street environment". "Mugspot" is the name of a new software module developed at the University of Southern California and the University of Bochum.
"Until now," says Christoph von der Malsburg, a professor at the University of Southern California and one of the system developers, "software for assigning faces to specific people required images in a very structured form: if possible, a sharp photo of a person, looking directly at the camera.”
The new "Mugspot" software looks out for passers-by when analyzing video images. If you find a corresponding pattern, it is selected and followed up as long as it is in the camera's field of view. After this "tracking" process, the software tries to find the most suitable image - the image in which the observer is looking as directly as possible into the camera. The module is capable of recording eight video frames per second in real time.
This image is then forwarded directly to the actual image processing software. The project described here involves facial recognition software that is already being used commercially by customers such as Deutsche Bank. The program has proven to be robust enough to be able to make identifications even on the basis of poor-quality recordings. It can even handle such difficulties as changing hairstyles, mustaches or full beards, and even sunglasses.
Up until now, the software needed about 13 seconds to select the best image, compare it with several hundred stored images and identify it. However, the developers hope to be able to reduce this time by half.
The Heidelberger Verlag Spektrum der Wissenschaft is the operator of this portal. Its online and print magazines, including "Spektrum der Wissenschaft", "Gehirn&Geist" and "Spektrum – Die Woche", report on current research findings.