Face Detection Using Classifiers Cascade Based on Vector Angle Measure and Multi-Modal Representation

This paper deals with face detection in still gray level images which is the first step in many automatic systems like video surveillance, face recognition, and images data base management. We propose a new face detection method using a classifiers cascade, each of which is based on a vector angle similarity measure between the investigated window and the face and nonface representatives (centroids). The latter are obtained using a clustering algorithm based on the same measure within the current training data sets, namely the low confidence classified samples at the previous stage of the cascade. First experiment results on refereed face data test sets are very satisfactory.