The Technology Research And Algorithm Of 3D Face Recognition

With the development of modern society, the biometric authentication such as face recognition, iris recognition, fingerprint recognition has attracted more and more attention. As an important aspect of biometric authentication, face recognition became one of the focus problems of academic research.Since 1990’s, the development of face recognition has been more than 20 years. Since the 3D data is hard to retrieve, the earlier face recognition is focused on 2D face recognition based on 2D images, and the research emphasis is focused on natural expressions and simple lights. As the development of research, face recognition has achieved satisfactory results under restricted conditions, but face recognition under complex conditions such as complex expressions, lights, postures gets little progress. At the same time, the development of 3D scanner machine makes 3D data easier to retrieve, so the 3D face recognition becomes the hot academic research. The 2D images are simple projections of 3D data essentially, so the 3D data gets more information than 2D image and 3D face recognition has potential to achieve bettor results than 2D face recongition.In this paper, the related algorithms of 3D face recognition are introduced first. Then the system flow of 3D face recognition is presented such as noise reduction, detection of nose point, extraction of face region and other pretreatment algorithms, the algorithm of pose correction, SFR, extraction of expression insensive region, face identification and verification. The main work and innovation of this paper are presented:1. We improved SFR with different distance mehtods. The SFR is improved with mahalanobis distance and standard euclidean distance. The classification rates are upgraded.2. A novel method of nose point detection based on face detecion is raised. We detect the face region with haar-like features first, then the horizontal slice is intersected with circle, and the intersections are used for decision of nose point. This method can eliminate most influence of clothes, hair, which may be detected as a nose point.3. A novel extraction method of expression insensive region is raised. This mehtod sets the horizontal and vertical range based on the nose point and extracts the expression insensive regions based on the range.4. A fully automatic system of face recognition is presented. The offline section is composed of pretreatment and extraction of SFR features. The online seciton is composed of SFR rejection, ICP and so on.5. Postures are adjusted with hotelling transform. The hotelling transform can adjust the face to the directions of most data whicn will eliminate most errors of ICP.