摘要
提出了一种表征曲面上某点局部形状的新方法———曲率符号分布形状描述子(CurSignTration).根据高斯曲率和平均曲率的符号确定曲面上点的形状分类,对某点邻域内所有点的形状按编码进行统计,将统计结果归一化后得到的直方图作为该点的CurSignTration.CurSignTration描述了三维深度图像中某点邻域的形状分布状况,对给定的三维深度数据点建立了一种除坐标之外的特征信息.这种表征方法对平移、旋转及镜像变换具有不变性.结合基于外观的人脸特征检测方法,用CurSignTration可以对人脸的三维深度图像进行姿态不变的脸部特征检测.对15张三维深度人脸图像的特征点检测结果表明了CurSignTration的有效性.
A new descriptor representing 3D local shape, called CurSignTration, was presented. For a point on surface, the CurSignTration is a histogram constructed from the statistical results of shape indices of the surface's points in the neighborhood of this point. The shape index was determined by the signs of Gaussian and mean curvature. The CurSignTration served to describe the structural neighborhood of a point in range data and establish a signature for each of the given 3D data points rather than the position information. It yielded invariance under rigid motions and mirror imaging. Its application in detecting facial features from range data was implemented by incorporating CurSignTration into the general framework of appearance-based face detection. The experimental results with 15 face range images showed the method's high accuracy in facial feature detection.
出处
《浙江大学学报（工学版）》
EI
CAS
CSCD
北大核心
2005年第5期652-656,共5页
Journal of Zhejiang University：Engineering Science
基金
国家自然科学基金资助项目(60273059)
浙江省自然科学基金青年科技人才培养专项基金资助项目(RC01058).
