ABSTRACT
A face recognition task can be viewed in general, as a combination of two stages, i.e., the face detection or verification stage and the face identification stage. The face detection algorithms usually include a verification phase at the decision level to discriminate the face and non-face portions of an image. These types of face detection problems can be tackled as a classification problem, where the given query image is classified into a face image (if a face is present in it) or a non-face image (if no face is present in it). For the purpose of classification, generally one needs to have samples from both the classes, and hence, as a training set, collections of face images and of non face images are required. The basic problem in this setup is to have information about all face
Face
and non face images. Division of the face space into subclasses can be done using a framework of higher order statistics to model the face and non-face clusters. Methods for face detection were also proposed [114] which seek to represent the wider variety of human faces as a set of subclasses.
