Detecting faces and recognizing facial features using color segmentation and 2DPCA in the normalized RG space

Face detection and recognition is very challenging due to the diverse variation of face appearance, facial expressions, variable recording conditions (changes in illumination, scale differences, varying face position...) and the complexity of image background. In this paper, we propose a new system which integrate color segmentation and two dimensional principal component analysis in the normalized RG space (2DPCA, to compress the red information as well as the green information) in order to detect faces and recognize facial features in color images that have not been preprocessed. We show some experimental results, using our own face database and the AR and PICS face databases. Then, we have compared results obtained with 2DPCA technique in the normalized RG space and other typical methods (2DPCA in the gray level space, PCA, Fisherfaces, Kernel PCA and Kernel Fisherfaces). Conclusions and future works have finally presented.