1. Subdivision/Refinement Dr. S.M. Malaek Assistant: M. Younesi

2. Introduction Bezier curves, B-spline curves and subdivision curves are all based upon the input of a control polygon and the specification of an algorithmic method that contructs a curve from this sequence of points. Fundamental to these methods is the concept of a refinement.

3. What is a Refinement Scheme?

4. A refinement process is a scheme which defines a sequence of control polygons where for any k>0, each can be written as:

5. What is a Refinement Scheme? For each fixed j and k the sequence is frequently called a mask. It covers the cases where the number of control points in each successive polygon is allowed to increase (Chaikin´s Curves and Doo-Sabin´s subdivision surfaces are examples of this), or must decrease (de Casteljau's algorithm for generating Bézier Curves is an example of this).

6. A Matrix Method for Refinement The equation can be written in matrix form as

7. A Matrix Method for Refinement (cont.) and overall Where S k is the refinement matrix This matrix as being sparse (i.e. most of the entries being zero) with non-zero entries clustered along the diagonal.

8. SUBDIVISION CURVES

9. A new set of curve generation methods is now becoming popular which utilize a control polygon, as in the Bézier or B-spline case, but instead of using analytic methods to directly calculate points on the curve, these methods successively refine the control polygons into an sequence of control polygons that, in the limit, converge to a curve. The curves are commonly called subdivision curves as the methods are based upon the binary subdivision of the uniform B-spline curves.

10. CHAIKIN'S ALGORITHMS FOR CURVES

11. CHAIKIN'S ALGORITHMS In 1974, George Chaikin gave a lecture at the University of Utah in which he specified a novel procedure for generating curves from a limited number of points. This algorithm is interesting as it was one of the first corner cutting or refinement algorithms specified to generate a curve from a set of control points, or control polygon.

12. CHAIKIN'S ALGORITHMS corner cutting or refinement algorithms: the algorithm generates a new control polygon by cutting the corners off the original one.

13. CHAIKIN'S ALGORITHMS Given a control polygon,we refine this control polygon by generating a new sequence of control points: where each new pair of points Q i, R i are to be taken taken to be at a ratio of and between the endpoints of the line segment.

14. CHAIKIN'S ALGORITHMS These 2n new points can be considered a new control polygon - a refinement of the original control polygon.

15. CHAIKIN'S ALGORITHMS Example: How Chaikin's Algorithm works? consider the following control polygon:

16. CHAIKIN'S ALGORITHMS Chaikin's algorithm generates the points Q i and R i and uses these points to refine the curve and obtain the control polygon shown in the figure below

17. CHAIKIN'S ALGORITHMS These points are in turn utilized to generate a new refinement.

18. CHAIKIN'S ALGORITHMS The following illustration shows the initial control polygon, the third control polygon in the sequence, and the final curve.

19. Subdivision Curves Example: A Closed Curves consider the following figure

20. Subdivision Curves We can still apply Chaikin's method to this figure, obtaining

21. Subdivision Curves This new control polygon can then be utilize to obtain a second refinement as in the following figure

22. Subdivision Curves The initial control polygon and the second refinement are shown in the following figure along with the resulting curve.

23. Subdivision Methods for Quadratic B-Spline Curves

24. Given a set of control points P 0,P 1, P 2.the quadratic uniform B-spline curve : For and where

25. Subdivision Methods for Quadratic B-Spline Curves We can perform a binary subdivision of the curve, by applying one of two splitting matrices to the control polygon: When applied to the control polygon S L gives the first half of the curve, and S R gives the second half. several of the control points for the two subdivided components are the same. Thus, we can combine these matrices

26. Subdivision Methods for Quadratic B-Spline Curves apply it to a control polygon i.e. at and along each of the lines of the control polygon. These are the same points as are developed in Chaikin's method.

27. Subdivision Methods for Quadratic B-Spline Curves The General Refinement Procedure: Given a control polygon, we can generate a refinement of this set of points by constructing new points along each edge of the original polygon at a distance of and between the endpoints of the edge. The general idea behind subdivision curves is to assemble these points into a new control polygon which can then be used as input to another refinement operation, generating a new set of points and another control polygon - and then continue this process until a refinement is reached that accurately represents the curve to a desired resolution.

28. Subdivision Methods for Cubic B-Spline Curves

29. Given a set of control points P 0,P 1, P 2 and P 3.The Cubic uniform B-Spline curve : For and where

30. Subdivision Methods for Cubic B-Spline Curves We can perform a binary subdivision by applying one of the two splitting matrices. We can combine these matrices

31. Subdivision Methods for Cubic B-Spline Curves Apply it to a control polygon

32. Subdivision Methods for Cubic B-Spline Curves The General Refinement Procedure In the case of the quadratic curve we were able to state exactly a single procedure for the points of the refinement. In the case of the cubic curve, it is not so easy. However, if we examine the rows of matrix used in the refinement, we see that they have two distinct forms. This motivates us to classify the points of the refinement as vertex and edge points. This classification makes the description of the refinement process quite straightforward.

33. VERTEX POINTS AND EDGE POINTS

34. Classifying the Refinement Points Suppose we are given a control polygon

35. VERTEX POINTS AND EDGE POINTS we use the refinement method: We note that three of the new control points appear to lie at the midpoints of the three respective line segments. These points will be classified as ``edge points''. The other points all lie close to on of the vertices of the original control polygon, and will be called ``vertex points''.

36. VERTEX POINTS AND EDGE POINTS we denote the new control polygon generated by the binary subdivision method as By combining and applying the splitting matrices that generate the binary subdivision of the curve we obtain

37. VERTEX POINTS AND EDGE POINTS The edge points E 0, E 1 and E 2 are calculated by:

38. VERTEX POINTS AND EDGE POINTS The V 0 and V 1 are calculated by:

39. VERTEX POINTS AND EDGE POINTS An Example of the Refinement Algorithm

40. VERTEX POINTS AND EDGE POINTS Calculate the edge and vertex points producing a refinement of the original control polygon

41. VERTEX POINTS AND EDGE POINTS Assemble these points into a new control polygon and utilize it as input again to the refinement process - producing new edge and vertex points from this set of points.

42. VERTEX POINTS AND EDGE POINTS Summary: In the case of cubic uniform B-spline curves we can consider the refinement procedure to consist of the generation of edge points and vertex points from a given set of control points - each of which can be generated by taking only the midpoints of line segments.

43. Subdivision Surfaces

44. Subdivision surfaces are based upon the binary subdivision of the uniform B-spline curve/surface. In general, they are defined by a initial polygonal mesh, along with a subdivision (or refinement) operation which, given a polygonal mesh, will generate a new mesh that has a greater number of polygonal elements, and is hopefully ``closer'' to some resulting surface. By repetitively applying the subdivision procedure to the initial mesh, we generate a sequence of meshes that (hopefully) converges to a resulting surface.

45. Subdivision Methods for Quadratic B-Spline Surfaces

46. Consider the biquadratic uniform B-spline surface P(u,v) defined by the array of control points M is the matrix:

47. Subdivision Methods for Quadratic B-Spline Surfaces This patch can be subdivided into four subpatches, which can be generated from 16 unique sub-control points. We focus on the subdivision scheme for only one of the four (the subpatch corresponding to ), as the others will follow by symmetry.

48. Subdivision Methods for Quadratic B-Spline Surfaces we consider the reparameterization of the surface by and and define this new surface as

49. Subdivision Methods for Quadratic B-Spline Surfaces Through this process, we have written the surface as The matrix S is typically called the “splitting matrix”

50. Subdivision Methods for Quadratic B-Spline Surfaces

51. where the can be written as

52. Subdivision Methods for Quadratic B-Spline Surfaces These equations can be looked at in two ways: 1.Each of these points utilizes the four points on a certain face of the rectangular mesh, and calculates a new point by weighing the four points in the ratio of 9-3-3-1. Thus, this algorithm can be specified by using subdivision masks, which specify the ratios of the points on a face to generate the new points. In this case, the subdivision masks are as follows

53. Subdivision Methods for Quadratic B-Spline Surfaces 2.Each of these equations is built from weighing the points on an edge in the ratio of 3-1. and then weighing the resulting points in the ratio 3-1. These are exactly the ratios of Chaikin's curve and so this method can be looked upon as a extension of Chaikin's curve to surfaces.

54. Doo-Sabin Surfaces

55. Subdivision Methods for Quadratic B-Spline Surfaces Given the subdivision masks generated for subdivision of biquadratic uniform B-spline patches Doo and Sabin observed that the points are simply the average of four particular points taken in a polygon - the vertex for which the new point is being defined, the two edge points (the midpoints of the edges that are adjacent to this vertex in the polygon), and the face point (average of the vertices of the polygon).

56. Subdivision Methods for Quadratic B-Spline Surfaces To verify this gives the subdivision masks, let the 4 vertex of a face be named A (bottom-right), B, C, D (upper-left) The 2 edge points adjacent to A are respectively equal and Finally the new point of the Doo- Sabin subdivision is equal to: The face point is equal to: A D C B

57. Subdivision Methods for Quadratic B-Spline Surfaces The Procedure for Meshes of Arbitrary Topology For each vertex of each face of the object, generate a new point as average of the vertex, the two edge points and the face point of the face.

58. Subdivision Methods for Quadratic B-Spline Surfaces Then for each face, connect the new points that have been generated for each vertex of the face

59. Subdivision Methods for Quadratic B-Spline Surfaces For each vertex, connect the new points that have been generated for the faces that are adjacent to this vertex. And then for each edge, connect the new points that have been generated for the faces that are adjacent to this edge.

60. Subdivision Methods for Quadratic B-Spline Surfaces Three iterations of a Doo-Sabin bicycle seat After the first subdivision, all vertices have valence four - which is a characteristic of the Doo- Sabin method.

61. Subdivision Methods for Cubic B-Spline Surfaces

62. Consider the bicubic uniform B-spline surface P(u,v) defined by the array of control points M is the matrix:

63. Subdivision Methods for Cubic B-Spline Surfaces The bicubic uniform B-spline patch can be subdivided into four subpatches, which can be generated from 25 unique sub-control points. We focus on the subdivision scheme for only one of the four (the subpatch corresponding to ), as the others will follow by symmetry. It should be noted that the nine ``interior'' control points are utilized by each of the four subpatch components of the subdivision.

64. Subdivision Methods for Cubic B-Spline Surfaces we consider the reparameterization of the surface by and and define this new surface as

65. Subdivision Methods for Cubic B-Spline Surfaces And we obtain:

66. Subdivision Methods for Cubic B-Spline Surfaces And we obtain (cont.):

67. Subdivision Methods for Cubic B-Spline Surfaces These equations can be looked in terms of subdivision masks as follows The application of mask A generates a new point corresponding to each vertex. The masks B generate new points corresponding to each edge, and mask C generates a new point corresponding to each face. These new points can be classified into three categories ­vertex points, edge points and face points ­depending on each points relationship to the original control point mesh.

68. CATMULL-CLARK SURFACES

69. Utilizing the subdivision for bicubic uniform B-spline surfaces, Ed Catmull and Jim Clark, following the methodology of Doo and Sabin noted that the subdivision rules expressed for the cubic B-spline surface not only work for arbitrary rectangular meshes, but can also be extended to meshes of an arbitrary topology. The Catmull­Clark subdivision process generalizes subdivision masks by calculating the face points, edge points and vertex points from an arbitrary mesh of control point to generate the subdivided control mesh.

70. CATMULL-CLARK SURFACES The points P’ 0,0, P’ 0,2, P’ 2,0 and P’ 2,2, which are shown in the figure below are called face points noted F i,j, and are calculated as the average of the four points that bound the respective face.

71. CATMULL-CLARK SURFACES We obtain: F 0,0 = P’ 0,0, F 0,1 = P’ 0,2, F 1,0 = P’ 2,0 and F 1,1 = P’ 2,2 Note that the face points are simply the average of the 4 points which define the face. These face points F i,j can be calculted by: F i,j =¼ (P i,j +P i+1,j +P i,j+1 +P i+1,j+1 ).

72. CATMULL-CLARK SURFACES The points P’ 0,1, P’ 0,3, P’ 1,0 P’ 1,2, P’ 2,1, P’ 2,3, P’ 3,0 and P’ 3,2, which are called edge points noted E i,j, are given as the average of 4 points ­the two points that define the original edge and the two new face points of the faces sharing the edge.

73. CATMULL-CLARK SURFACES The example of the 2 edge points E 0,1 and E 1,0 P’ 0,1 can be rewrited as follow E 0,1 = P’ 0,1 =¼( (P 0,0 +P 1,0 +P 0,1 +P 1,1 )/4 + P 0,1 + P 1,1 + (P 0,1 +P 1,1 +P 0,2 +P 1,2 )/4)) which is also equal to ¼ (F 0,0 +F 0,1 +P 0,1 +P 1,1 ). So this kind of edge points E i,j can be calculated as follow E i,j =¼ (F i,j-1 +F i,j +P i,j +P i+1,j ) P’ 1,0 can be rewrited as follow E 1,0 = P’ 1,0 =¼( (P 0,0 +P 1,0 +P 0,1 +P 1,1 )/4 + P 1,0 + P 1,1 + (P 1,0 +P 1,1 +P 2,0 +P 2, 1)/4)) which is also equal to ¼ (F 0,0 +F 1,0 +P 1,0 +P 1,1 ). So this kind of edge points E i,j can be calculated as follow E i,j =¼ (F i-1, j +F i,j +P i,j +P i,j+1 )

74. CATMULL-CLARK SURFACES Note that the edge points are the average of 4 points : the 2 face points of the 2 faces adjacent to the edge and the 2 points which define the edge. These edge points E i,j can be calculted either as E i,j =¼ (F i,j-1 +F i,j +P i,j +P i+1,j ) Or E i,j =¼ (F i-1,j +F i,j +P i,j +P i,j+1 )

75. CATMULL-CLARK SURFACES The remaining four points P’ 1,1, P’ 1,3, P’ 3,1 and P’ 3,3, as shown in the figure below are called vertex points noted V i,j.

76. CATMULL-CLARK SURFACES Let’s take the example of the vertex point V1,1 P’ 1,1 can be rewrited as follow (1) V 1,1 = P’ 1,1 = 1/16 ( (F 0,0 +F 1,0 +F 0,1 +F 1,1 ) + (P 1,0 +P 0,1 +P 2,1 +P 1,2 ) + 8 P 1,1 ) Note that the vertex points are then the average of the 4 face points adjacent to the vertex point, with the 4 endpoints of the 4 edges adjacent to the vertex point and with 8 times the corresponding vertex from the original mesh.

77. CATMULL-CLARK SURFACES By denoting Q = 1/4 (F 0,0 +F 1,0 +F 0,1 +F 1,1 ), R = 1/4 ( (P 0,0 +P 1,1 )/2 + (P 1,0 +P 1,1 )/2 + (P 2,1 +P 1,1 )/2 + (P1,2+P1,1)/2) and S = P 1,1, we obtain V 1,1 = P’ 1,1 = 1/4 (Q + 1/4 (P 0,0 +P 1,0 +P 2,1 +P 1,2 ) + P 1,1 -P 1,1 + 2 P 1,1 ) = 1/4 (Q + 2R -P 1,1 + 2 P 1,1 ) = 1/4 (Q + 2R + P 1,1 ) = 1/4 (Q + 2R + S) (2)

78. CATMULL-CLARK SURFACES Note that the vertex points are also equal to 1/4 (Q + 2R + S) with Q the average of the 4 face points adjacent to the vertex point, R the average of midpoints of the 4 edges incident on the vertex point and S is the corresponding vertex from the original mesh. Depending on the reference, formula (1) or formula (2) is used to define vertex points.

79. CATMULL-CLARK SURFACES Subdividing Meshes of Arbitrary Topology Given a mesh of control points with an arbitrary topology, we can generalize the face point, edge point, vertex point specification from the uniform cubic B­ spline surface calculations to obtain

80. CATMULL-CLARK SURFACES For each face of the mesh, generate the new face points ­which are the average of all the original points defining the face (We note that faces may have 3, 4, 5, or more points now defining them). Generate the new edge points ­which are calculated as the average of the two points which define the original edge with the two new face points of the faces adjacent to the edge. Calculate the new vertex points ­which is equal to 1/n (Q + 2R + (n- 3)S), where n is the order of the vertex, and where Q is the average of the new face points of all faces adjacent to the original face point, R is the average of the midpoints of all original edges incident on the original vertex point, and S is the original vertex point. This calculation rule of the vertex point generates surfaces that exhibit tangent plane continuity at all extraordinary points. With the rule 1/4 (Q + 2R + S) it was observed that in some arbitrary meshes, tangent plane continuity was not maintained at extraordinary points.

81. CATMULL-CLARK SURFACES Four steps in the Calmull-Clark refinement process

82. CATMULL-CLARK SURFACES

84. Summarize Doo­Sabin and Catmull­Clark Subdivisions Surfaces

85. Doo­Sabin (respectively Catmull­Clark) surface is locally a bi­ quadratic (respectively bi­cubic) B­spline surface, except at a finite number of control points (extraordinary points). Even at those points the Catmull­Clark surface satisfies tangent plane continuity. Both subdivision schemes are uniform and stationary approximating scheme mostly for quadrilateral net but have rules for subdividing n­sided polygons. The Doo­Sabin (respectively Catmull­Clark) refinement rule is based on corner cutting – dual (respectively vertex insertion ­primal). After the first subdivision step of Doo­Sabin scheme, all vertices have valence four. After a single subdivision step of Catmull­Clark scheme, all the faces in the control net are quadrilaterals.