Research Of Face Recognition Technique Under Variable Lighting And Pose

Automatic face recognition has great potential applications and great theory value in many subjects. Over the past 30 years, great progress and developments have been made in face recognition. However, under the uncontrolled and uncooperative conditions, especially when the illumination in face images and facial poses are variant, the recognition rate degrades quickly and face recognition is also a great challenge.On the analysis of the present methods for these two problems, a novel method which handles both pose and lighting conditions is proposed here via this paper. First, the intensity of images in training set is normalized in order to reduce the susceptivity toward lighting intensity. Afterwards, a low dimensional illumination space representation (LDISR) of human faces for arbitrary lighting conditions is proposed. The 5 basis images under the 5 basis point light sources construct a LDISR, during which the images in training set can be estimated. Then the conception called"closest illumination ratio image (CIRI)"is proposed, which is designed to reconstruct face images under normative illumination and expand training set. The images'pose is also estimated, and eigenvector subspace of different pose is built up by means of eigenface. Ultimately, the conception called"pose's weight value (PWV)"is proposed to normalized Euclidian distance. Due to this approach, the problem of variant pose was solved by assigning different pose's weight value.Experimental results on FERET and Yale B database demonstrate that this technique could significantly improve the accuracy of face recognition under variant illumination and pose conditions. Moreover, compared with the resent methods, the novel one need less pattern images, while the training speed and recognition speed is faster, and the restrictions on the images are much less. Therefore, the proposed method has great potential applications.