Principal Investigator:

Investigators:

The human visual system can recognize unprimed views of common objects at sustained rates in excess of 10 per second (between 5.000 and 30.000 different objects!). A common assumption is that this process is based on a feedforward feature extraction hierarchy. From a computational perspective this assumption leads to several interesting questions: How to select the best set of features? Can this features be learned from examples? Which kind of classifiers are suited for this kind of architectures? How to build robust models from local features?