具有拓扑鲁棒性的三维人脸识别算法研究

项目名称： 具有拓扑鲁棒性的三维人脸识别算法研究

项目编号： No.61202268

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 计算机科学学科

项目作者： 孙粤辉

作者单位： 广东工业大学

项目金额： 23万元

中文摘要： 本项目针对三维人脸识别中影响准确性的主要因素- - 表情变化与拓扑噪声干扰进行研究，旨在建立更具准确性与鲁棒性的三维人脸识别算法。本项目主要研究内容是： 1、欧式距离与测地距离融合为散布距离,用以表征曲面上点的平均距离而非最短距离（测地距离），将三维人脸转化成散布距离构成的度量空间(Metric Space)，在此度量空间中计算相似度,通过外形匹配来完成识别任务，不仅使人脸识别对表情变化不敏感，也不受几何外形拓扑结构改变及三维数据获取所引入的拓扑噪声的影响，对原有的三维人脸识别算法理论做突破性改进。 2、提出使用双树复小波(DT-CWT)直接对三维数据进行处理，提取人脸更具区分性的纹理特征，并与拓扑鲁棒性的散布距离结合，进一步提高识别准确性。

中文关键词： 三维人脸识别；散布距离；双树复小波变换；；

英文摘要： The goal of this research project is proposing a more accuracy and robustness algorithm to avoid the influence of expression changes and topological noise on 3D face recogntion. The main contents of this project are: 1. Diffusion distance is established by combining the Euclidean distance and the geodesic distance, by using which the surface points' connections are characterized with the average distance rather than the shortest distance ( geodesic distance ). 3D face is modeled mathematically as a metric space whose all elements are the fusion distances. The recognition task is completed like shape matching by computing the similarity of the metric space.The algorithm is not sensitive to expression changes, geometry topology changes and also topological noise during 3D data acquisition which result the breakthough improvment on 3D face recognition. 2 The 3D data is processed directly by the Dual Tree Complex Wavelet Transform(DT - CWT) to extrac more distinguishable features of the texture. The performance is further improved by using the diffusion distance joined in the DT CWT features.

英文关键词： 3D Face Recognition；Diffusion Distance；Dual Tree-Complex Wavelet Transform；；