【作者】 史利民；

【导师】 王仁宏；

【作者基本信息】 大连理工大学 ， 计算数学， 2005， 硕士

【摘要】 基于大规模散乱数据的插值或拟合方法,在很多领域都有重要的应用。所以长期以来,都有很多学者从事这方面的研究,并且发展和形成了许多方法。本文主要是针对其中的若干常用方法,提出一些改进办法。同时结合实例比较,证明改进后的算法在计算复杂度或拟合质量等方面较原来的方法有了较大的提高。第一章主要改进了Shepard插值方法。对于常用的Shepard方法,一般地认为,其权函数的光滑性和衰减性越好,曲面的拟合效果就越好。该章结合截断多项式、B样条基函数和指数函数来构造其权函数,使新的权函数具有更高的光滑度和更好的衰减性,并且其光滑性和衰减性可以根据实际需要自由调节。从而提高了曲面的拟合质量。通过大量的实例表明,运用改进的方法得到的曲面确实比用原先的算法得到的曲面无论从光滑性和光顺性上都有了较大的提高。同时该方法还可用于修正的Shepard方法等。第二章主要给出了一个基于径向基(非局部支集)的局部插值方法。用径向基(非局部支集)插值,随着型值点数量的增加,其系数矩阵的阶数以及条件数也迅速增大,从而给计算带来不便。在本章中,结合多元样条的思想,给出了一个使径向基插值局部化的算法。该算法较好地继承了径向基插值曲面的性质,从而保证了拟合曲面具有好地光顺性和拟合精度。曲面片之间的光滑性可以灵活选择。算法构造简单,计算方便,具有较强的实用性。文中数值实例也表明运用分片局部化算法,在计算效率上有了很大的提高,尤其是在型值点数量较大的情况下。而曲面拟合质量与直接用径向基插值得到的曲面具有可比性。第三章主要用参数曲面-NURBS拟合散乱数据。该章首先推广了文献的结果,将文献中关于B样条曲线曲面拟合数据点的迭代算法推广至有理形式,给出了无需求解方程组反求控制点即可得到拟合NURBS曲线曲面的迭代方法。该算法和文献的算法本质上是统一的,而后者恰是前者的一种退化形式。同时本章还给出了收敛性证明以及一些定性分析。另外结合前两章的内容,对文献提出的用NURBS曲面拟合散乱数据的抽样网格方法作了适当的改进,提高了算法运算效率,并对曲面拟合质量也有一定的提高。同时还提出一种基于散乱数据的均匀网格抽样方法。由于均匀网格适宜参数化,所以更便于NURBS拟合。本文运用了大量实例来验证算法,都收到了比较好的效果。更多还原

【Abstract】 The problem of constructing approximations based upon scattered data are encountered in many areas of scientific applications, like meteorological information, such as the amount of rainfall, or geological information, such as depth of underground formmations. This is done using interpolation techniques that estimate value on unexplored points of a region considering the values sampled on it. The main work of this thesis is to improve some approximation methods that are often used in practice. With the numerical examples, we prove the improved methods .are more convenient to compute or have more approximate quality.In chapter 1, with regard to the Shepard method, we use the truncated polynomials, the B-spline basis functions and exponential functions to construct the weight functions. They are of better properties of smoothness and decay, which can be adjusted freely. And so the surface can be fitted better by the improved method. It also can be applied in modified Shepard Mehthod[5].In practice, interpolation methods of radial basis functions are often required for approximation with very large number of data, in which the interpolation matrix is usually ill-conditioned. In order to solve this problem, we construct a localized method for interpolation with radial basis functions (global supported) based on the idea of multivariate spline. Moreover, we verify that this method is feasible through theoretic analysis and numerical experiments. These methods perform almost as good as the global supported radial basis functions interpolation methods do.In chapter 3, we use NURBS to approximate the scattered data. Paper[20]gave out the iterative algorithm of B-spline interpolation and approximation. In this chapter,we generalize this result and present an iterative algorithm of NURBS interpolation and approximation. Using this algorithm, we can get the approximat NURBS curve or surface directly without solving a linear system to compute the weights and control points. This algorithm is consistent with the algorithm in paper [20] and the latter is just the degenerate form of the former in essence.Furthermore, using the new methods introduced in former chapters, we improve the NURBS approximation method given out in paper[21], and then the computing is simpler. We also present an approach for scattered sampling and uniform mesh generation. Since the uniform mesh is easy to be parameterized.this approach is more convenient to be used. .The numerical examples in this thesis show us these methods are feasible, and the results are satisfying.更多还原