Smooth parametric surface approximations to discrete data (Q1067752)

In this paper are given the well known methods of \textit{J. L. Earnshaw} and \textit{I. M. Yuille} [A method of fitting parametric equations for curves and surfaces to sets of points defining them approximately, Comput. Aided Des. 3, 19-22 (1971)], the method of \textit{J. A. P. Kjellander} [Smoothing of cubic parametric splines, ibid. 15, 175-179 (1983) and Smoothing of bicubic parametric surfaces, ibid. 15, 288-293 (1983)] and the SSR1 Sculptured Surfaces system. The author suggests that there are two important points to be noted in parametric approximating surfaces. Firstly, all residuals are measured in the same sense, i.e. in the z-direction (specified x and y values are usually taken to be exact). Secondly, there is a natural ordering of points in both the x and the y direction. Consider first the residuals. One reason for the use of parametric surface in engineering is that they are largely axis- independent. It's therefore inappropriate to measure residuals in a particular direction. The natural course is to take the residual at each data point to be the normal distance to the surface. Next consider the ordering of data points. In the parametric case the ordering must be made in a curvilinear u,v coordinate system embedded in the parametric surface. But if we commence with nothing but a set of data points the surface doesn't yet exist. On the other hand, we cannot assign u,v values to the data points until a surface has been defined. The paper raises a number of questions which could form the basis for useful research in this area.