ESCA Workshop on Audio-Visual Speech Processing (AVSP'97)
September 26-27, 1997
Our work addresses the localization of faces in the setting of a relatively smooth background and under uncontrolled illumination conditions, and in relatively large indoor scenes. Our approach is based in some simple heuristics applied to the vertical and horizontal projections of the binarization of the gradient magnitudes extracted by a Sobel operator from the smoothed image. Experimental results on a sequence of images demonstrate the robustness of our approach. Our approach provides a fast and robust preprocessing that can be used by many systems that process facial information for sophisticated human-computer interaction, including speechreading systems.
Bibliographic reference. Raducanu, B. / Gratia, M. (1997): "An approach to face localization based on signature analysis", In AVSP-1997, 109-112.