摘要：The generation of models with high degree of realism has been possible through the advance of equipments and techniques for spatial data acquisition. Many applications require massive volumes of data, such as computer vision, medicine, remote sensing and virtual reality. Triangle meshes are data representations with various advantages over the use of regular grids, including adaptability to data density, ease of manipulation and visualization of complex surfaces, and organization of structures at di.erent levels of resolution. This paper describes a method for constructing triangle meshes from images at multiple scales smoothed with Gaussian ˉlters. A new metric for incrementally inserting data points into the mesh is proposed, which is robust in the presence of noise or outliers. Experimental results demonstrate that the proposed approach generates compact meshes while maintaining the original data surface approximation at a proper level of accuracy..
