3D Face Recognition Based On Global Feature And Scale Invariant Feature

Recent years, 3D face recognition has become one of the most active and potential fields in pattern recognition and machine vision. As a kind of biometric technology, it has a wide range of applications.3D face recognition is based on 3D data. Compared with 2D face recognition, 3D face feature can adequately represent human face. Also, 3D face recognition not only contains rich information but also can overcome the effects of illumination, expression, pose, makeup etc. The procedures of 3D face recognition are: acquisition and pre-processing of 3D data, feature extraction, feature matching.To solve the key problems of 3D face recognition, the research work of this thesis is as follows:(1) Due to the face models in 3D face database contain the non-face information, such as shoulder, ear, hair etc, choose the tip of the nose as the reference point and a right radius to get the accurate face area. Then, use the depth value of facial cloud points as pixel value to get the depth image of face area. And get the gray image of face area by the RGB value of the facial cloud points.(2) Due to the large computation of the Scale Invariant Feature Transform(SIFT) algorithm, an improved SIFT algorithm is proposed. The improved SIFT algorithm divide the neighborhood of a SIFT feature key point into nine seed regions, generate a 72-dimensional feature descriptor. Compared to the original SIFT algorithm, the improved SIFT algorithm can effectively reduce the computational complexity when feature matching.(3) To extract the effective facial features as the basis of recognition process, Principal Component Analysis(PCA) algorithm is used to get the facial global feature from the 2.5D depth image, and an improved SIFT algorithm is used to get the facial local invariant feature from the gray image.(4)Based on global feature and local invariant feature of 3D face, a two-layer classifiers is applied to recognize faces. The experiment results show that 3D face recognition based on both global feature and SIFT feature can achieve higher face recognition rate than using single feature. Also, the two-layer classifiers make the recognition process more efficient.