Construction and Recognization of 3D Faces

• Main Authors: Chang Jui Lun, 章瑞倫
• Other Authors: 郝樹聲
• Format: Others
• Language: zh-TW
• Published: 2011
• Online Access: http://ndltd.ncl.edu.tw/handle/43806478094189278518

碩士 === 國防大學理工學院 === 電子工程碩士班 === 99 === In this thesis, we try to take the advantages of 3D inherent nature in face recognition. The 3D face recognition is not restricted in its view angles because the inherent nature of stereo image. We use two images taken from different angles to construct a 3D face model. With this model we can further process to make facial features recognition. The 3D models are constructed and encoded by the structure light in our research with our own built 3D camera system. In recognizing, we try to full utilize the curvatures of nose and mouth because they are invariant in different facial expressions. These 3D curvatures can be used as features to recognize. The Gaussian curvatures and mean curvatures are adopted in our thesis. The distribution of these two curvatures is used to build the feature vectors. The re-mesh and normalization methods are also applied to the 3D mesh models. After the curvatures have been extracted, we apply the Iterative Closest Point (ICP) algorithm to make the comparison. In order to raise the recognition rate, the projection method is used to intensify the edge information too. Finally, the 3D normalized curvature feature distance is measured between the query and database images. The correctness of reorganization rate is reached approximately to 90%.