3D Implicit Surfaces Reconstruction Algorithm Based On Improved RBF Network

This paper studies and discusses the implicit surface recon-struction from three-dimensional scattered point cloud, which is based onradialbasisfunctiondata. Forlocalimplicitsurfacereconstructionmethod,this paper completed the following tasks:Firstly,forthepresenceofemptydatamodel,radialsurfacereconstruc-tion method based global multi-scale can effectively repair voids. Howev-er, this method is part of a global approach, that is, when the data is toolarge the matrix will betoo big to calculate, so beyond the existing comput-ing level. To be able to scale the big data, topology complexity, and withempty model, we propose a multi-scale radial basis partial surface recon-struction methods. The method combined with the strengths of multi-scalemethods to repair voids and local method of high efficiency. The final re-construction results are strongly illustrating the realization of the purpose.Then, we consider the sample size is large, when a visual needs ofreconstruction is confirmed, the data maybe redundant in fact for the p-resence of the model. Based on this consideration, we introduce l1sparseregularization theory into surface reconstruction problem, which means,add a l1norm constraint into all the interpolated data points, making thesolved coefficient obtained is sparsity. Based on the obtained coefficientvalues, those values corresponding to the deleted data points is small,that is the contribution that these data points can be ignored for reconstruc-tion function, a large value means large contribution to the retention of datapoints. Finally, experimental results show that this method compared withthe general data sampling methods, effectively retained the local featuresof the model. In addition to the noise data, this method can delete noisepriority, which is robust to noise. Finally, the traditional method of surface reconstruction, generally re-quire additional constraints of off-set points, which undoubtedly greatlyreduces the effect of surface reconstruction, and the increase in empiri-cal constraints to the model also brings instability. In order to avoid theadditional constraints from the surface point, we propose a local surfacereconstruction methods which is displayed through coordinate transforma-tion. weassumethatthelocalsurfacesheetcanusealocalexplicitfunctionto approximate, then we convert the unkown explicit function into its pre-sentation of other coordinate system. Experimental results show that themethod can handle surface model with different characteristics; comparedwith the traditional method based on radial basis of surface reconstruction,the time is shortened by about2/3; compared MPU method Ohtake pro-posed is more effective resistant to noise.