Research On Features Extraction And Face Recognition From Range Image

Face recognition is a kind of biometric technology, which is one of the most active fields in pattern recognition and machine vision. Furthermore, it has many potential applications. By means of extracting and analyzing the facial feature, human recognition and authentification are achieved. Generally, it deals with 2-D image, and has attained a series of improvements. However, 2D images can change dramatically due to pose, illumination, expression and viewing variations, which might bring problems to recognition. In order to solve these problems, the international experts begin to analysis the 3-D range images, and have developed several relational methods. Yet the current researches are quite insufficient.In terms of 3-D human feature, the thesis analyzes and researches the methods of feature extraction and recognition. Firstly, an overview of face recognition, especially the 3-D face recognition, is presented. Then the facial description is devised from the point of psychophysicists. Thereafter, the algorithm is put forward to detect and label 3-D facial feature automatically and the strategy is studied to search and recognize 3-D samples from Database, too. Afterwards, the relationship between 3-D human feature and 2-D images (frontal and profile) is analyzed, and retrieval and recognition using 2-D sample image are carried out. Finally, the analysis, design of the system and programming are presented, and then the developing tendency and application prospect of the 3-D face recognition technology are expected. The main works are as follows:1) The hierarchical describing method of facial pattern is presented, the facial feature is quantitative evaluated by the Euler distance of feature points, curvature feature of profile and subjection degree of organ, and the biometric facial coordinates is defined, then the relationship between 3-D human feature and 2-D images is researched..2) By means of the experiential knowledge, the automatic detection and label of the 3-D human feature is achieved. And using the method of radial difference and improved Gaussian image, the facial edge region is obtained. Based on the template, the accurate feature points are extracted by the algorithms of fitting region edge and optimizing energy function.3) The method to verify the extraction result based on double-layer semantic nets is advanced, and the results of detecting 3-D human feature automatically are arranged to construct the semantic nets, which is validated in terms of knowledge rules.4) Making use of 3-D samples, the face recognition is implemented by the Euler distance feature, the profile subsection curvature and the outline membership degree, while the multiple classifiers is combined by improved voting method.5) The method using the 2-D photo to retrieve and recognize the 3-D sample is brought forward, which combines the convenience for obtaining the 2-D photo and the stability of 3-D samples feature, so expands the application fields of 3-D face recognition technology.