Reconstruction of sub-sampling surface based on support vector machine

In computer graphics and reverse engineering, 3D laser scanners are becoming standard devices for input data, providing for millions of points. As a consequence, point-base geometric models generating from unorganized point clouds has gained more and more attention. And points are primitives of surface modeling and rendering. The advantage of point-based techniques is that point sets do not require any connectivity information, which significantly reduces computation cost of re resentations and rendering of complex geometric models. Point clouds representations are discrete samplings of a given object and hence proper reconstruction approaches have to be applied in order to enable hole-free rendering. Surface splatting is an effective surface reconstructing algorithm, which uses circular or elliptical surface splats to cover point primitives. But the computation costs are still proportional to the number of surface splats used to represent a geometric model. In this paper, a statistical learning techniques based sub-sampling algorithm was proposed for surface splatting modeling, point cloud data preprocessing techniques which is based on epsilon-Support Vector Regression Machine (epsilon-SVRM) and nu-Support Vector Regression Machine (nu-SVRM) was put forward as well. This algorithm proceeds in three steps, firstly, surface profile is constructed by epsilon-SVRM. secondly, surface fitting is applied by nu-SVRM. finally, greedy optimization algorithm is used to obtaining optimized splat-based surface representations of geometric models. The experiment results indicate that the proposed approach can generate smooth splat-based surface within given error tolerance with fast processing speed.