3D Face Recognition Based On Global Feature And Local Feature

Face recognition is regarded as an important research area in biometric identification, have attracted much research institute’s attention. Because the3D point clouds describe the accurate geometric feature of the object and are invariant to the person’s posture and illumination, the research of3D face recognition has become more and more popular. The feature of the face can be grouped into two categories, the global feature and the local feature. The global feature’s calculation is very simple, but it’s influenced notable by the person’s posture, lighting and expression. The local feature is contrary to the global feature, and it’s invariant to posture, lighting and expression. This paper proposes a novel face recognition based on global feature and local feature, combining their respective advantages.First, this paper introduces the procedure of3D face point cloud’s preprocessing, and converts it to2.5D depth image and intensity image. Then, we extract the PCA global feature and SIFT local feature from the depth image and insensitive image. The global feature represents the overall information of the face, and the local feature represents the local information invariant to posture, lighting and expressions. So this paper’s algorithm uses the global feature to reject the most faces, remaining the candidate face according to the global similarity. Lastly, we use the SIFT local feature to do the accurate recognition. The numerical experiments show that this paper’s algorithm perform well than the face recognition based on single feature, and the global feature speed up the whole face recognition system significantly.