Pre-Confirmation Neural Network for Reducing the Region of Interest in an Image for Face Detection

Abstract

In this paper we present a Pre-Confirmation Neural Network (PCNN) to reduce the processing time of the general face detection Neural Networks (NNs) by reducing the region of interest in an image up for face detection. The other algorithms commonly used for most face detection works by applying one or more NNs, directly to portions of the input image, and arbitrating their results. This requires that the whole image be passed several times through different NNs thereby increasing the processing time required for face detection. We present a smaller and less complex PCNN which operates on the image to produce a relatively small set of image portions which have the possibility of being a Face. When only this small set is passed through the NNs, generally used for face detection, the time required to detect faces in an image reduces. © Springer-Verlag Berlin Heidelberg 2010.