Advances in Component Based Face Detection

We describe the design and construction of a component based face detector for gray scale images. We will show that including parts of the face into the negative training sets of the component classifiers leads to improved system performance. We introduce a method of using pairwise position statistics between component locations to more accurately locate the parts of a face. Finally, we illustrate an application of this technology in the creation of an accurate eye detection system. Thesis Supervisor: Tomaso Poggio Title: Uncas and Helen Whitaker Professor of Brain and Cognitive Sciences