Research On The Key Issues Of Facial Recognition System

As one of biometric, human's face has characteristic such as: exclusive, easy to gain, can be remote sensing. Thus, facial recognition is a wildly researched task of biometric recognition. This paper begins at intellective facial recognition system and studies on the problems: facial detection, facial expression recognition, pose-emendation, illumination's affection what facial recognition system has counted with.1, Facial location: the arithmetic that projects the facial complexion model to YCrCb color-space is mainly analyses. This method is robust when there is positive human face in the picture.2, Facial expression recognition: the enhanced facial expression sub space arithmetic is analyses. Facial expression sub-space is gain by using PCA arithmetic, and then projects the unclassified picture into every facial expression sub-space; we can get the coordinate of facial expression.3, Pose-emendation: the methods using KPCA and KDDA are mainly analyses. Kernel function is applied in PCA and DDA, so the linear discriminance method is dislinearized.4, Illumination's affection: homomorphic filtering is used as the pretreatment of PCA to reduce affection of illumination. The rate of correct recognition is improved by 10 percent. Illumination Cone is studied. Three facial picture under different illumination condition in which there is a facial picture under standard illumination are used to build illumination cone. The pictures of face under different illumination and the 3d surface of human face can be gain by using illumination cone. In order to reduce the affection of illumination, color-invariants method is used to transact multicolor facial picture. The pictures with similar histogram are got.In conclusion, facial recognition is an important study subject. This paper studies on the key issues of intellective facial recognition system: facial detection, facial expression recognition, pose-emendation, illumination's affection. The expressional picture is unnecessary to identify by using the method based on enhanced human facial expression sub-space; the information of facial 3d surface can be used in facial recognition after using illumination cone method; multicolor pictures of human face can be transacted by using the method based on color-invariants.