Expression Invariant 3D Face Recognition Based On Diffusion Distance

As capture point cloud data more conveniently, the research based on point cloud data attracts more and more researchers. The attribute of 3D point cloud data which is not sensitive to illumination and position makes it superior to 2D image data. Thus,3D face recognition has gradually become a hot area of face recognition, In 3D face recognition, expression variance is the main factor that influences the rate of recognition accuracy. At past, obtaining the features of 3D facial expression variance is based on geodesic distance, but usually it is very complex on computation and has low rate of recognition accuracy on open mouth expression. Based on that, our paper use diffusion distance as the metric of isometric transformation and propose a novel algorithm of obtaining expression invariant features.The main content of this paper includes:first, introduce the common geometry invariance in pattern recognition, and focus on the mathematical principal of construction and computation of geometric invariance under isometric transformation. Then, based on geometric invariance under isometric transformation, we proposed the theory framework of facial expression invariant 3D face recognition. Specific work includes:First, we get depth and intensity images from the aligned 3D data. Then, we compute the diffusion distance from all the points to the nose tip in depth image, thus, distance image and level curves can be generated. Further, we can obtain invariant features vectors from intensity values in intensity image at fixed points. At last, recognition work is completed using invariant feature vectors by LDA and Mahalanobis distance. Our experiment result shows that our algorithm performs better than the one based on geodesic distance.