形状空间下的3D面貌相似性比较

开展了利用计算机自动比较两个3D面貌相似程度的研究。提出了一种在形状空间下基于测地线的3D面貌相似性比较的方法。该方法使用测地线近似表示3D面貌模型,然后将其变换到形状空间下,以对应测地线的平均测地距离作为两个3D面貌的相似性比较的依据。在公开的GAVADB三维面貌库上对14个人的28个3D面貌模型进行了实验,识别率达到92.86%以上,说明本方法可以将同一个人的不同模型与不同人的模型区分开。另外,使用提出的方法对不同人的3D面貌模型进行了相似性比较实验,结果表明使用本方法得出的相似程度的判断与人的主观判断相一致,说明本方法能够较好地反应3D面貌的相似程度。

3D facial similarity comparison in shape space

This paper focuses on the 3D facial similarity comparison by using a computer automatically. A new 3D facial similarity comparison method based on a geodesics in a shape space was proposed. In this method, 3D facial surface was approximately represented by a set of geodesics and the corresponding geodesics between two facial models were compared by a elastic metric in the shape space, in which the similarity between two faces could be measured by the average geodesic distance of all corresponding geodesics. 3D facial similarity experiments were perfomed on 28 facial models of 14 persons in Gavadb database and the recognition rate is above 92.86%, which indicates that this method recognizes the facial models of the same person from different person's facial models. In addition, 3D facial similarity experiments on different person's facial models were implemented. The experimental results show that the similarity values computed by proposed method consist with that of people's subjective judgment. These results reflect that the proposed method can measure the similarity of 3D faces well.