Summary: E cient Focusing and Face Detection
Yali Amit, Donald Geman y and Bruno Jedynak z
Technical Report no. 459
Department of Statistics
University of Chicago
We present an algorithm for shape detection and apply it to frontal views of faces in still
grey level images with arbitrary backgrounds. Detection is done in two stages: (i) \focusing,"
during which a relatively small number of regions-of-interest are identi ed, minimizing com-
putation and false negatives at the (temporary) expense of false positives; and (ii) \intensive
classi cation," during which a selected region-of-interest is labeled face or background based
on multiple decision trees and normalized data. In contrast to most detection algorithms, the
processing is then very highly concentrated in the regions near faces and near false positives.
Focusing is based on spatial arrangements of edge fragments. We rst de ne an enormous
family of these, all invariant over a wide range of photometric and geometric transformations.
Then, using only examples of faces, we select particular arrangements which are more common in
faces than in general backgrounds. The second phase is texture-based; we recursively partition
a training set consisting of registered and standardized regions-of-interest of both faces and