Adaptive multiresolution representation of a point-based 3D model for rendering

A multiresolution representation of a 3D model is one of important means for its fast rendering. In this paper we present an approach to build up an adaptive multiresolution representation for a point-sampled 3D model directly from its set of sample points. The binary space partition (BSP) is used to recursively split a point-sampled model according to the covariance analysis of its sample points. Thus a model is subdivided along the direction of the greatest variation, and the partition is adaptive to the spatial distribution of sample points. A multiresolution hierarchy based on bounding spheres is constructed for a model from its subdivision. Since a point-sampled model has not any topological information, the k-nearest clustering method is adopted to find a local neighborhood for each sample point, and a surface normal at each sample point is estimated by fitting a tangent plane in its neighborhood. We construct adaptive multiresolution hierarchies for several point-sampled models and display them based on the point-based rendering. Experimental results show this approach is efficient for rendering complex point-based 3D models.