3D Face Recognition Based On Sparse Representation

As capture point cloud data more conveniently, more and more scholars to research the3D face recognition. Compared with the2D image data,3D point cloud data contain more information, and the attribute of3D point cloud data which can better handle changes in illumination, pose and expression makes it superior to2D image data. Thus,3D face recognition has gradually become a hot area of face recognition. In3D face recognition, it is often complex computationally on the feature extraction because of the relatively complex facial information, and the result is also not very ideal. Coupled with the changes in illumination, pose and expression, the recognition accuracy decreased significantly. Sparse representation with fewer features can effectively express the important information of the3D face. Using it to face recognition, we can get the ideal recognition rate, and the feature extraction method is no longer critical and the computation and processing complex relatively reduced. Based on that, our paper use sparse representation to the3D face recognition, and the experiments on the actual database verify the advantages of the method having a small amount of calculation, ideal and stability of recognition rates.The main content of this paper includes:first, we introduce commonly used feature extraction methods in face recognition, and focus on the theory and algorithm of the extraction method based on the principal component analysis and the geodesic distance. Then, the theoretical framework of the3D face recognition system based on sparse representation is given. Specific work includes:first, we get depth images from the aligned3D data; then, we compute the feature vector; further, the recognition work is completed using sparse representation. Our experiment result shows that the recognition method based on sparse representation in3D face recognition can get an ideal recognition rate, and feature extraction method is not sensitive, with relatively fast computing speed.