Abstract: The traditional method of facial similarity is easily affected by the subjective consciousness and the limitation for semi-automatic extracting the key features is inaccurate, the paper puts forward a 3D face similarity measurement based on adaptive neighborhood description. Firstly, features points are marked automatically based on profile of faces model after the faces are unified to Frankfurt coordinate. Then, the feature description is constructed by the covariance matrix of the geometric feature with the adaptive neighborhood. Finally, the similarity between description of feature points illustrates the similarity between two face model. Experimental result shows that the proposed method can measure the similarity of face model well, and distinguish the similarity of face models.

Key words: 3D facial similarity, feature points, geometric invariant, feature descriptor

摘要： 针对传统面貌相似性度量方法容易受主观影响以及半自动面貌特征点提取方法存在标定不准确等问题，提出一种基于自适应邻域特征描述子的三维面貌相似性度量的方法。该方法将待比较面貌模型统一到法兰克福坐标系下，并利用等测地线方法根据面貌模型侧面轮廓线自动提取特征点；通过构建特征点的自适应邻域几何特征协方差矩阵，作为特征点描述子；对特征点的特征描述子进行相似性度量从而确定面貌模型间的相似性。实验结果表明，该方法不仅可以有效地评判三维面貌模型的相似性，而且能够区分相似面貌和不相似面貌模型。

关键词: 三维面貌相似性, 特征点, 几何不变量, 特征描述子