Face recognition from color images in presence of dynamic orientations and illumination conditions

Abstract

Current face recognition systems work well in certain controlled conditions with frontal images. However, they do not take some unpredictable factors into consideration. For example, in dynamic environments with complex backgrounds and textures, or when the illumination or contrast is relatively low, most methods fail to perform accurate feature recognition. In this paper, we propose a robust face recognition system based on color reference model, efficient light bias method, face morphing and independent component analysis. This system performs face recognition against complex dynamic backgrounds and from variable angles of view. Our experimental results show recognition rates of over 92% using an optimized ICA algorithm. © Springer-Verlag Berlin Heidelberg 2004.