Please activate JavaScript!
Please install Adobe Flash Player, click here for download

uni'wissen 02-2012 ENG

Observed from different perspectives: Videos provide the computer with a wealth of image data that show the variations of an object. This allows it to learn, for instance, what a sheep looks like – and to acquire the ability to find sheep on other pictures. When we see dachshunds, Dobermans, or Great Danes, we know right away: They’re all dogs – no matter what perspective we see them from and no matter whether they are eating, sleeping, or fetching a stick. Computers do not have this knowledge. For them pictures are noth- ing other than a long chain of numbers composed of points with various color values. Prof. Dr. Thomas Brox from the Department of Computer Science of the University of Freiburg wants to teach computers to grasp the meaning of these abstract data: “The goal is an automatic learning process in which computers find objects on pic- tures and improve their search independently.” In order for a computer to recognize objects, it has to be able to come to grips with their varia- tions on two different levels. To take the example Recognition 2.0 12