A pre-filter enabling fast frontal face detection

Abstract

We present a novel pre-filtering technique that identifies probable frontal illuminated face regions in colour images regardless of translation, orientation, and scale. The face candidate regions are normalised and provide the basis for face verification using published face detection algorithms. The technique focuses on a fast search strategy to locate potential eye-pairs in an image or video frame. The eye-pair candidates indicate areas that may contain faces. Scale and orientation is inferred from the eye-pairs, and a neural network is used to confirm the normalised face candidates.