Research Of 3D Face Recognition Based On Extended LBP Features

As an important branch of biometric recognition technology, face recognition technology has been extensively applied in access control system, authentication and public security. Compared to 2D face recognition,3D face recognition has pose and illumination invariance which has attracted more and more attention and becomes a hot topic in current field of face recognition. However,3D face data includes more shape information, making the acquisition more difficult than 2D image. Besides, how to select the useful information is a key problem in 3D face recognition. At the same time,3D face recognition is also affected by expression variations, so how to overcome the influence of facial expressions is an important research direction of 3D face recognition. To solve the problems above, the main work of this paper is listed below:1) A 3D face recognition method based on extended local binary pattern of region is proposed. First, the preprocessing is preformed on the faces in the gallery and probe, including face cutting, smoothing and denoising pointcloud and face pose normalization. Then, the depth image converted from the preprocessed 3D pointclouds is normalized. Rigid region, semi-rigid region and non-rigid region according to their distortions under facial expressions are extracted by binary masks and represented by the uniform pattern of the extended LBP. Finally, Score-level fusion with weighted SRC(W-SRC) of rigid region and semi-rigid region are also tested. The experiments on FRGC v2.0 database demonstrate that the proposed method is robust and efficient.2) A 3D recognition method based on extended meshLBP is proposed. Interest points of rigid region are extracted and the local regions around these points are also defined. Then, to use the 3D information of face model more effectively, the histogram of extended meshLBP in every local region is extracted according to the ordered ring defined in the meshLBP. Through feature-level fusion the extended meshLBP descriptor of the whole face is obtained. Finally, the experiments with minimum distance classifier are conducted and demonstrate the proposed algorithm is robust to expression variations.3) A 3D face recognition system is realized which is based on the research of the 3D face recognition technology above and combined with the 3D measurement system developed in our laboratory. The procedure of 3D face recognition includes face data preprocessing, numerical feature extraction, central profile extraction and accurate matching. Combing rapid rejection algorithm with accurate matching algorithm, the system has high recognition rate and fast recognition speed.