Polyhedral design with blended n n -sided interpolants

Péter Salvi · 2026

Read Polyhedral design with blended n n -sided interpolants by Péter Salvi (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2601-19322.txt.

Introduction

Péter Salvi designs smooth surfaces that pass through every corner of a given mesh. A closed mesh of any topology sets the corners; the surface must interpolate them all. His method blends local patches, each a multi sided quadratic piece, into quadrilateral patches that join smoothly.

Faces with other than four sides need a special parameterization built from rational curves. Triangles among the subpatches get their own handling, with alternate forms discussed. Smooth through point surfaces serve design and animation, where a cage of points must become a skin that looks unbroken.

Catalog record

CREATORPéter Salvi
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE14 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2601-19322

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Polyhedral design with blended n n -sided interpolants Péter Salvi Abstract A new parametric surface representation is proposed that interpolates the vertices of a given closed mesh of arbitrary topology. Smoothly connecting quadrilateral patches are created by blending local, multi-sided quadratic interpolants. In the non-four-sided case, this requires a special parameterization technique involving rational curves. Appropriate handling of triangular subpatches and alternative subpatch representations are also discussed. 1 Introduction Surface representations based on control polyhedra come in various guises, including recursive subdivision [ 1 ] , generalized splines [ 7 ] and ‘manifold’ approaches [ 5 ] . In this paper we examine a variation of the manifold-based construction exemplified by the works of Zorin [ 13 , 10 ] , and recently by Fang [ 3 ] . Figure 1: Overlapping quadratic nets around a quad. Figure 1 shows the basic idea. The input is a closed mesh. We assume that it only contains quads – if not, we can perform a central split on all faces (similarly to a Catmull–Clark subdivision step) to get rid of the multi-sided patches while retaining all original vertices. For each quad, the 1-rings around its corners define (multi-sided) quadratic control nets. We can generate four local patches interpolating these control points, and evaluate them in the parametric region associated with the quad (Fig. 2 ). The resulting surface is created by blending these together. Figure 2: Blending the local interpolants. Dots indicate a point of evaluation in each; the last image shows the resulting patch. Note that while most similar methods are approximative, here we focus on patches that interpolate the input points. The interpolation criterion makes the interpolants more similar in the given region, which may enhance the quality of the resulting surface. The rest of the paper is organized as follows. Section 2 shows the construction in the simple case of regular quadmeshes. Handling of irregular vertices is explained in Section 3 , with the definition of Quadratic Generalized Bézier (QGB) patches in Section 3.1 and a parameterization based on rational Bézier curves in Section 3.2 . Some test results are presented in Section 4 ; notes on future work conclude the paper. 2 Regular meshes In a regular (closed) mesh all vertices have a valency of 4, so all control nets around a quad will be quadrilateral. We use quadratic tensor-product Bézier patches for the interpolants. Let \mathbf{C}_{i} and \mathbf{E}_{i} ( i=1\dots 4 ) denote the corner and edge control points, with \mathbf{E}_{i} being between \mathbf{C}_{i-1} and \mathbf{C}_{i} (with cyclical indexing). The middle control point is denoted by \mathbf{M} . Then the interpolant is defined as \mathbf{I}(u,v)=\sum_{i=0}^{2}\sum_{j=0}^{2}\mathbf{P}_{ij}B_{i}^{2}(u)B_{j}^{2}(v), (1) where B_{k}^{d}(t) is the k -th Bernstein polynomial of degree d at parameter t , and \displaystyle\mathbf{P}_{00} \displaystyle=\mathbf{C}_{3}, \displaystyle\mathbf{P}_{10} \displaystyle=\hat{\mathbf{E}}_{4}, \displaystyle\mathbf{P}_{20} \displaystyle=\mathbf{C}_{4}, \displaystyle\mathbf{P}_{01} \displaystyle=\hat{\mathbf{E}}_{3}, \displaystyle\mathbf{P}_{11} \displaystyle=\frac{1}{4}(16\mathbf{M}-\sum_{i=1}^{4}(\mathbf{C}_{i}+2\hat{\mathbf{E}}_{i})), \displaystyle\mathbf{P}_{21} \displaystyle=\hat{\mathbf{E}}_{1}, \displaystyle\mathbf{P}_{02} \displaystyle=\mathbf{C}_{2}, \displaystyle\mathbf{P}_{12} \displaystyle=\hat{\mathbf{E}}_{2}, \displaystyle\mathbf{P}_{22} \displaystyle=\mathbf{C}_{1}. (2) Here \hat{\mathbf{E}}_{i}=2\mathbf{E}_{i}-\frac{1}{2}(\mathbf{C}_{i-1}+\mathbf{C}_{i}) is the control point position s.t. [\mathbf{C}_{i-1},\hat{\mathbf{E}}_{i},\mathbf{C}_{i}] define a quadratic Bézier curve interpolating \mathbf{E}_{i} . The parameterization of the quad is in [0,1]^{2} , and is rotated locally with the following interpolation constraints in mind ( \mathbf{S} denotes the patch to be created): \displaystyle\mathbf{S}(0,0) \displaystyle=\mathbf{M}, \displaystyle\mathbf{S}(1,0) \displaystyle=\mathbf{E}_{1}, \displaystyle\mathbf{S}(0,1) \displaystyle=\mathbf{E}_{2}, \displaystyle\mathbf{S}(1,1) \displaystyle=\mathbf{C}_{1}. (3) The interpolant point corresponding to the (u,v) point in the quad’s domain is then \left((u+1)/2,(v+1)/2\right) . Consequently, the quad patch can be defined as the blend of the four interpolants: \mathbf{S}(u,v)=\sum_{i=1}^{4}\mathbf{I}_{i}\left(\frac{u_{i}+1}{2},\frac{v_{i}+1}{2}\right)\Phi(u_{i},v_{i}). (4) The local parameterizations are defined as \displaystyle u_{1} \displaystyle=u, \displaystyle v_{1} \displaystyle=v, \displaystyle u_{2} \displaystyle=v, \displaystyle v_{2} \displaystyle=1-u, \displaystyle u_{3} \displaystyle=1-u, \displaystyle v_{3} \displaystyle=1-v, \displaystyle u_{4} \displaystyle=1-v, \displaystyle v_{4} \displaystyle=u. (5) The blending function is the product of two 1-argument blends \Phi(u,v)=\Psi(u)\cdot\Psi(v) (6) with the following constraints: \displaystyle\Psi(0) \displaystyle=1, \displaystyle\Psi(1) \displaystyle=0, \displaystyle\Psi^{(k)}(0) \displaystyle=\Psi^{(k)}(1)=0 (7) for some k>0 . For finite k , Hermite blends can be applied: \Psi(t)=\sum_{i=0}^{k}B_{i}^{2k+1}(t). (8) For k=\infty there are several options, including bump functions and expo-rational B-splines [ 9 ] . We have used the k=2 Hermite blend for all examples in the paper. 3 Irregular vertices An irregular vertex generates a non-quadrilateral control net. We interpolate these points by quadratic generalized Bézier patches, as defined below. 3.1 Quadratic Generalized Bézier (QGB) interpolants (a) Bilinear map (b) Discontinuity (c) Rational curve map (d) Triangle case Figure 3: Quad \rightarrow interpolant mappings. Generalized Bézier (GB) patches [ 11 ] are normally defined only for cubic or higher degrees, but is easy to further generalize the construction to the quadratic case. As this will be a C^{0} interpolant, its equation will be even simpler than that of the original. The surface is defined over the regular n -sided polygon \{(\cos\frac{2k\pi}{n},\sin\frac{2k\pi}{n})\} , k=0\dots n-1 , and for each side we create local coordinate mappings \displaystyle s_{i}(u,v) \displaystyle=\lambda_{i}/(\lambda_{i-1}+\lambda_{i}), \displaystyle d_{i}(u,v) \displaystyle=1-\lambda_{i-1}-\lambda_{i}, (9) where \{\lambda_{i}\} are the Wachspress coordinates [ 4 ] of (u,v) relative to the domain polygon. Using the notations of the previous section, \displaystyle\mathbf{I}(u,v) \displaystyle=\sum_{i=1}^{n}\left(\mathbf{C}_{i-1}\frac{1}{2}B_{0}^{2}(s_{i})+\hat{\mathbf{E}}_{i}B_{1}^{2}(s_{i})+\mathbf{C}_{i}\frac{1}{2}B_{2}^{2}(s_{i})\right)B_{0}^{2}(d_{i}) \displaystyle\qquad\quad\,+\mathbf{P}_{0}B_{0}(u,v). (10) Here B_{0} denotes the weight deficiency B_{0}(u,v)=1-\sum_{i=1}^{n}\left(\frac{1}{2}B_{0}^{2}(s_{i})+B_{1}^{2}(s_{i})+\frac{1}{2}B_{2}^{2}(s_{i})\right)B_{0}^{2}(d_{i}), (11) and the central control point \mathbf{P}_{0} is defined s.t. the patch interpolates \mathbf{M} : \mathbf{P}_{0}=\frac{n^{2}\mathbf{M}-\sum_{i=1}^{n}\left(\mathbf{C}_{i}+2\hat{\mathbf{E}}_{i}\right)}{n(n-3)}. (12) Note that for n=4 this is exactly the same as the tensor-product interpolant in Section 2 . It is also easy to see that for n=3 this is a quadratic triangular Bézier patch. 3.2 Parameterization We still need a mapping between the quad domain [0,1]^{2} and the interpolant’s domain (a regular polygon inscribed in the unit circle). The associated part of the latter is a kite defined by the points (0,0) , (c,-s) , (1,0) and (c,s) , where c=\frac{1}{2}(\cos(2\pi/n)+1) and s=\frac{1}{2}\sin(2\pi/n) , see Figure 3 . A naïve approach would be to use a bilinear mapping (Fig. 3a ). This presents a problem however: with this simple mapping adjacent parts of the same interpolant would be parameterized discontinuously (Fig. 3b ). In the rest of this section, we will construct a mapping that is C^{\infty} -continuous (except at the origin, which is singular). The proposed method works by creating pencils of quadratic rational Bézier curves in the two parametric directions, and the mapping of (u,v) is defined to be the intersection of the u -curve with the v -curve – see Figure 3c , where the following notations are also shown: \displaystyle\mathbf{V}_{0} \displaystyle=(c+\hat{c},-s-\hat{s}), \displaystyle\mathbf{V}_{1} \displaystyle=(2c-1,-2s), \displaystyle\mathbf{V}_{2} \displaystyle=(c,-s), \displaystyle\mathbf{V}_{3} \displaystyle=(1,0), \displaystyle\mathbf{V}_{4} \displaystyle=(c,s), \displaystyle\mathbf{V}_{5} \displaystyle=(2c-1,2s), \displaystyle\mathbf{V}_{6} \displaystyle=(c+\hat{c},s+\hat{s}), (13) where \hat{c}=\frac{1}{2}(\cos(4\pi/n)-1) and \hat{s}=\frac{1}{2}\sin(4\pi/n) . The control points and rational weights of u -curves are given as \displaystyle\mathbf{R}_{0} \displaystyle=\mathbf{V}_{0}(1-u)+\mathbf{V}_{1}u, \displaystyle w_{0} \displaystyle=1, \displaystyle\mathbf{R}_{1} \displaystyle=\mathbf{V}_{2}u\exp(u^{2}-u), \displaystyle w_{1} \displaystyle=1/u, \displaystyle\mathbf{R}_{2} \displaystyle=\mathbf{V}_{4}(1-u)+\mathbf{V}_{3}u, \displaystyle w_{2} \displaystyle=1. (14) Similarly for v -curves: \displaystyle\mathbf{R}_{0} \displaystyle=\mathbf{V}_{6}(1-v)+\mathbf{V}_{5}v, \displaystyle w_{0} \displaystyle=1, \displaystyle\mathbf{R}_{1} \displaystyle=\mathbf{V}_{4}v\exp(v^{2}-v), \displaystyle w_{1} \displaystyle=1/v, \displaystyle\mathbf{R}_{2} \displaystyle=\mathbf{V}_{2}(1-v)+\mathbf{V}_{3}v, \displaystyle w_{2} \displaystyle=1. (15) Then the curves themselves can be evaluated with the formula \mathbf{r}(t)=\frac{\sum_{i=0}^{2}\mathbf{R}_{i}w_{i}B_{i}^{2}(t)}{\sum_{i=0}^{2}w_{i}B_{i}^{2}(t)}. (16) The exponential term in the middle control point is there to accommodate for the triangular domain where the \mathbf{V}_{2}-\mathbf{V}_{3}-\mathbf{V}_{5}-\mathbf{V}_{6}-(0,0) polygon is concave, and this helps pulling back the curve near the origin. Figure 3d shows v -curves in a triangle for v=k/20 values, k=0\dots 20 . For a given (u,v) point in the quad domain, we need to intersect the corresponding u - and v -curves. This is easily done with nested golden section searches in the [\frac{1}{2},1] interval [ 6 ] . While this is a bit expensive, it can (and should) be precomputed.

Read the complete text