Face location by template matching with a quadratic discriminant function

We present the application of a new approach for object location to the problem of locating frontal and near-frontal views of faces in still grey-scale images. The approach uses a linear and a quadratic discriminant function in a hierarchical fashion to classify subimages of a fixed size extracted from the input image at various positions and resolutions. The parameters of each discriminant function are chosen such that they maximize an objective function which uses the first and second order statistics of the values of the discriminant function on face and non-face images. The resulting classifier is accurate and very fast, and is part of the commercial face recognition system FaceVACS.