Honeycomb Subdivision Ergun Akleman¤and Vinod Srinivasan Visualization Sciences Program, Texas A&M University Abstract ² Corner cutting schemes such as Doo-Sabin [8, 3]: af- ter one iteration all vertices 1 become 4-valent and the In this paper, we introduce a new subdivision scheme number of non-4-sided faces remains invariant after which we call honeycomb subdivision. After one iteration the first refinement. of the scheme each vertex becomes exactly 3-valent and ² Vertex insertion schemes such as Catmull-Clark [5]: with consecutive applications regular regions strongly re- after one iteration all faces become 4-sided and the sembles a honeycomb. This scheme can be considered as a number of non-4-valent vertices is constant after the dual for triangle schemes. The major advantage of the new first refinement. scheme is that it creates a natural looking mesh structure. p ² Triangular schemes such as 3-subdivision [10, 11]: 1. Introduction after one iteration all faces become 3-sided and the number of non-6-valent vertices is constant after the first refinement. Although subdivision surfaces were introduced more than 20 years ago by Doo, Sabin, Catmull and Clark[5, 8], Notice that in this list vertex insertion and corner cut- they were ignored by the computer graphics industry until ting schemes are dual, i.e., one of them makes every face they were used in 1998 Academy Award-winning short film 4-sided, the other one makes every vertex 4-valent. There “Geri’s Game” by Pixar [7, 20]. Since then subdivision is currently no dual for triangular schemes. The scheme we surfaces have become increasingly popular in the computer present in this paper provides the missing dual for triangular graphics and modeling industry. This is not a surprise since schemes: subdivision methods solve the fundamental problem of ten- ² Honeycomb schemes: after one iteration all vertices sor product parametric surfaces [12, 13] without sacrificing become 3-valent and the number of non-6-sided faces the speed of shape computation [5, 8, 11, 9, 18, 7]. Unlike is constant after first refinement. tensor product surfaces, with subdivision surfaces, control meshes do not have to have a regular rectangular structure. We call such schemes honeycomb since the resulting Subdivision algorithms can smooth any 2-manifold (or 2- meshes strongly resemble honeycombs, which our dictio- manifold with boundary) mesh [20, 21]. nary defines as (1) A structure of hexagonal, thin-walled Subdivision surfaces assume that users first provide an ir- cells constructed from beeswax by honeybees to hold honey regular 2-manifold control mesh, M0. By applying a set of and larvae, (2) Something resembling this structure in con- subdivision rules, a sequence of finer and finer 2-manifold figuration or pattern. meshes M1; M2;:::; Mn;::: are created. These meshes Figure 1 shows five iterations of our honeycomb scheme. eventually converge to a ”smooth” limit surface M1 [10]. In this example, the control mesh M0 is a dodecahedron, One way to classify subdivision schemes is based on M1 is a truncated icosahedron or soccer ball [19]. More what kind of regularity emerges with the application of the interestingly, the mesh strucures from M2 to M5 can be scheme [16]. In other words, the pattern of regular regions found in virus structures [17]. As seen in this example, the can be used to characterize the scheme. Based on regular most important property of the new scheme is that the re- regions, existing subdivision schemes can be classified into sulting mesh structures strongly resemble natural structures three major categories: such as cells or honeycombs. It is also interesting to note that their structure looks similar to Voronoi diagrams. ¤Corresponding Author: Address: 216 Langford Center, College Sta- tion, Texas 77843-3137. email: [email protected]. phone: +(979) 845- 1Vertices and faces are also called vertets and facets in order to avoid 6599. fax: +(979) 845-4491. confusion with the vertices and faces of a solid model [16]. 0 The remainder of this paper is organized as follows. In vertex vn (See Figure 2.B) (we assume that the ver- the next section, we provide refinement rules for our general tex and edge indices are given in the order of a face honeycomb scheme. In Section 3, we introduce the coeffi- traversing). Compute the position of each new ver- cients of our honeycomb algorithm that are used to create tex as a linear combination of the old vertex positions 0 the meshes shown in Figure 1. Section 4 explains how to fv0; v1; : : : ; vn; : : : ; vN¡1g. Note that since vn’s cor- implement a remeshing algorithm for honeycomb subdivi- respond to edges instead of vertices, in order to derive sion. Section 5 discusses implementation and results. Fi- the coefficients of a honeycomb scheme, it will be use- nally, our conclusion is given in Section 6. ful to use the following linear equation: NX¡1 0 ¡ ¢ vn = an;m (1 ¡ t)vm + tv(m+1) (mod N) (1) m=0 where the an;m’s are real coefficients, N is the valence of the face and t is a real number between 0 and 1. (A) Initial mesh (B) New vertices are created (C) New faces are created Figure 1. Five iterations of our honeycomb scheme over a dodecahedron. (D) New edges are inserted (E) Final mesh Figure 2. Illustration of honeycomb process. 2. Refinement Rules for Honeycomb Schemes Remark 1. Note that (1 ¡ t)vm + tv(m+1) (mod N) is We give the refinement rules for a general honeycomb a point on the edge em. In our examples we generally algorithm as follows (The algorithm is also illustrated in use t = 0:5. Figure 2, where the black vertices and edges are new and ² Step 2 : For each face, create a new face by connecting constitute the new mesh, and the gray vertices and edges all the new points that have been generated by that face are in the old mesh and are removed by the algorithm.): (See Figure 2.C); ² Step 1 : For each edge en of a face f = ² Step 3 : For each edge of the mesh, connect the two fe0; e1; : : : ; en; : : : ; eN¡1g of the mesh, create a new new points that have been generated for that edge (See Figure 2.D); In these equations, the parameter a in equation (2) is provided by the users and is as a tension parameter [4]. ² Step 4 : Remove all old vertices and edges (See Fig- Note that the value of a controls the speed of convergence. ure 2.E). We suggest making a a decreasing function of N, i.e, for Remark 2. Let Fn, En and Vn denote the number of higher-sided faces (larger values of N) the algorithm must faces, edges and vertices created by nth iteration. It converge faster. In all the examples in this paper we use the is straightforward to show that Fn = Fn¡1 + Vn¡1, following function for a En = 3En¡1 and Vn = 2En¡1. Based on these equa- 1:25 tions, it follows that a = 0:45 + : N F lim n ¡! 3: n!1 Fn¡1 It is easy to verify that in this scheme, each coefficient an;m is always greater than zero if 1 > a > 0, and the co- In other words, this honeycomb subdivision scheme efficients do add up to 1. By using Fourier analysis, sim- increases the number of faces 3-fold at each itera- ilar to Doo-Sabin’s approach, we can show that the new tion. Therefore,p regular haneycomb scheme can also scheme also has three distinct but complex eigenvalues2. be called a 3 honeycomb algorithm [10]. The complex part comes from the phase shift introduced Remark 3. The support region of our honeycomb by the term vm + v(m+1) (mod ()N): Absolute values of the scheme (the region that is influenced by a given con- other eigenvectors will be smaller than one if a < 1, regard- trol vertex [16]) has a fractal boundary, a Koch island less of the value of N. Similar to Doo-Sabin’s scheme, only [15]. one eigenvector corresponds to the eigenvalue 1, two eigen- vectors correspond to the second largest (in absolute value) Remark 4. It is difficult to identify the precision eigenvalue and the rest of the eigenvectors correspond to the set of any honeycomb scheme (comparing the scheme smallest eigenvalue. Because of this property our scheme with a parameteric surface) since the regular regions of provides tangent plane continuity. the honeycomb scheme do not correspond correspond known to any parameterization scheme such as the ten- sor product or box-spline parameterizations. 4. Implementation and Results 3. The Coefficients of Our Honeycomb Algo- The remeshing algorithm explained in the section 2 has rithm been implemented in a C++ program. We applied the hon- eycomb algorithm to the various control shapes. Figure 3 Like any subdivision scheme relying on approximation, shows examples of applying our honeycomb scheme. The images at the left in Figure 3 show the control meshes and in honeycomb subdivision the coefficients an;m in step 1 in the above algorithm must satisfy the following conditions: images at the right show the mesh after four iterations of our honeycomb algorithm. It is interesting to note that the 1. an;m ¸ 0 for all n and m and honeycomb algorithm creates a very natural looking mesh structure. It therefore can be used after other subdivision N¡1 X schemes to create natural looking meshes. 2. for all n a = 1. n;m The honeycomb algorithm can create lateral artifacts m=0 [16]. Figure 4.(B) shows such an artifact.