三维曲面的特征描述和特征匹配

3D Surface Feature Description and Feature Matching

本文提出了一种新的三维曲面特征描述算法,将二维图像上的特征描述思想推广到三维网格。算法将三维网格表示成从顶点到高斯曲率的映射函数,从而获得可类比于二维图像的相似性。借助于法线和梯度建立局部球坐标系,通过二维统计直方图对特征点邻域的几何信息进行描述,使得特征描述具有平移、旋转和缩放不变性,最终生成128维的特征向量(特征描述符)。基于特征向量,我们实现了多分辨率和异拓扑网格下的特征匹配,展示并分析了实验结果。本文的研究动机来源于三维扫描建模以及多视点三维重建技术中对特征描述和特征匹配的需求,主要的应用方向包括:扫描配准、模型注册、动画跟踪、对称检测和模型检索。

In this paper,we propose a new algorithm of 3D surface description and extend the idea of feature description on 2D images to 3D meshes.The algorithm represents 3D meshes as a mapping function from vertices to Gaussian curvatures,so the similarity to 2D images is obtained.Equipped with normal and gradient,we build local spherical coordinate and describe the geometrical information around feature point by 2D histogram.The descriptor is invariant to the changes in translation,rotation and scale.Based on the finally produced feature vectors（feature descriptors） with 128 dimensions,we carry out the feature matching on meshes with multi-resolutions and different topologies.We present and analyze the experimental results.The paper is motivated by the need for feature description and feature matching,due to the recent advancements of 3D scan modeling and multiple camera 3D reconstruction.The main application includes scan alignment,model registration,animation tracking,symmetry detection and model retrieval.