3D Facial Feature Extraction And Recognition Based On Face Profile

Now 3D face recognition has been one of the most popular topics in the field of pattern recognition and image processing.In this paper we research the stepped approach for three-dimensional facial recognition in the combination of curvature and geometric features. Firstly, we obtain the 3D facial disparity by two 2D images which were taken with two parallel cameras. Then, we analyse the 3D data, standardize the 3D data by the grid control points, and then extract the 3D face features, finally do the 3D face recognition. Our work contributes to the following aspects:1. 3D facial data acquisition module: we introduce several methods of 3D facial data acquisition and introduce the principle of binocular vision, and then, we obtain the 3D facial disparity by two 2D images which were taken with two parallel cameras. In addition, we also analyse and obtain data by following research from 3D facial databases provided by other research institutions.Then, we standardize the 3D data: simulating the 3D model which is created by B-spline surface fitting, and using the grid control points to generate the point-cloud data, finally resample the 3D facial data.2. 3D facial feature extraction module: this paper made use of depth and other characteristics to extract three profiles of 3D face. After pre-treatment it accurately anchored ten facial feature points by implementing related curvature analysis and calculation and deciding the datum marks based on the combination of curvature metric space and curvature mark changes. This improves robustness of the algorithm.3. 3D facial recognition module: on the basis of anchoring feature points, we get the feature data by geometric calculation, which can be used as parameters for facial recognition. And then we do comparison between two methods of facial recognition by sub-curvature features of four profiles and ten feature points. One of the methods is based on geometric measuring. The other is based on curvature of profiles. After the comparison, we adopt stepped approach for three-dimensional facial recognition in the combination of curvature and geometric features, which has better performance by using weighted-mean of multiple features corresponding to multiple facial expressions and stepped approach.Do research on 20 models and total 140 faces , the Rank one has accury rate of 91 %.