Three Dimensional Face Recognition

Face recognition is one of the most popular research fields at present. Recently, most of the researches on face recognition are based on 2D face image. Because of the influence of illumination, pose variation and expression, the improvement of recognition accuracy of 2D face recognition is greatly impeded. This makes it still difficult to build a robust face recognition system. 3D model holds more rich information thatn 2D image, so implementing face recognition on 3D face model is one of the effective approaches to tackle the present problems.The thesis mainly discusses how to extract efficient and robust 3D face features for recognition after necessary preprocessing of 3D face models and how to reduce the computing loads of recognition, due to the large quantity of data and inherent high complexity of 3D model, with some preprocessing.The contribution of the thesis is summarized as follows:1. In the part of "Curvature Analysis and Estimation", several kinds of curvature feature are discussed in theory and analyzed with experimental results. An approach of "Interior Expansion" is proposed to tackle the problem of curvature computing on the border of 3D model. In addition, 3D model is smoothed based on the information of curvature and its integer quality is improved. During the process of model smoothing, there is an uncertain weight threshold, which is discussed in detail for scientific evaluating.2. In the part of "3D Model Segmentation", the vector of Gauss-Mean curvature, that of Max-Min curvature, that of RMS-Absolute curvature are compared through their performances in segmentation. The effect of cluster number is also tested through experiments. On request of 3D face recognition, segmented model should be post-processed so as that numerous small sub-regions can be merged into meaningful parts.3. In the part of "3D Model Registration", firstly essential regions relating with five organs on face are extracted under the constraints of spatial relation among five organs and area of regions on the model. The centroid of each essential region plays as landmark on the model and is made use of in the process of coarse registration. Different from the original landmark detection algorithm, this one isfree of influences of mode and scale of present face model.4. In the part of "3D Face Recognition", a new feature with global and local information of model combined with each other is proposed. Globally, the spatial relation within the essential regions relating with five organs is extracted as part of model feature. Besides, feature vector of every essential region, obtained by computing the mean feature vector of all the vertices in the region, is concatenated in decided order as local feature.