Geometric and Topological Multi-Resolution of n-dimensional Solids

Transcription

1 Geometric and Topological Multi-Resolution of n-dimensional Solids Luiz Velho IMPA - Instituto de Matematica Pura e Aplicada

2 Acknowledgements PhD. Thesis Work Esdras Medeiros Filho (IMPA) Co-Supervision Helio Lopes (PUC-Rio) Collaboration Thomas Lewiner (PUC-Rio) Luiz Velho October,

3 Outline Context and Motivation Mathematical Background Sampling Solid Objects Geometric / Topological Multiresolution Adaptive Multi-Triangulation Structure Examples and Applications Future Work Luiz Velho October,

4 Motivation Why Solids? Most Objects are NOT Hollow Matter Inside! [Kopf et al, 2007] An Empty Glass is Full of Air Inside Matters ;-) [C-SAFE, 2005] Luiz Velho October,

5 3D Applications Modeling CSG Animation Deformation Simulation [Romeiro et al, 2005] [de Goes et al, 2007] Fluids Scientific Visualization [de Goes et al, 2007] Luiz Velho October,

6 Image Analysis Segmentation 2D Applications [Felzenszwalb et al, 2004] Image Compression Adaptive Triangulation [Demaret et al, 2000] Medical Imaging Contouring [Lewiner et al, 2006] Luiz Velho October,

7 Related Work Geometric Multiresolution Multi-Triangulations Adaptive Meshes Etc.. [Marroquim et al, 2005] Topology Control Simplification Filtration / Persistence Etc.. [Zomorodian et al, 2005] Luiz Velho October,

8 What is Missing? Unified Geometric Topological Framework For Example: Simplest Topology / Highest Resolution Only Holes with Size Greater than X Fine Boundary and Coarse Interior * Why it is difficult? Continuous Geometry x Discrete Topology Representation / Data Structure Adaptation, Computation, etc [Romeiro et al, 2005] Luiz Velho October,

9 Game Plan Characteristic Function Multiresolution Hierachy Adaptation * Obs: Attributes! Luiz Velho October,

10 Relations with M.M. Morphological Scale-Spaces Mathematical Morphology Dilation / Erosion Binary Image (Characteristic Function) * Discrete Representation Level Sets Diffusion Interface Curve (Boundary) * Continuous Front Evolution Luiz Velho October,

11 Key Ideas Our Framework Solid Topology + Stochastic Point Sampling Geometric / Topological Operators Variable-Resolution Structure Advantages Integrate Geometry and Attributes Natural Notion of Scale General Adaptation Luiz Velho October,

12 Contributions Solid Poisson-Disk Sampling Scale Space for Characteristic Function α-filtration Multiresolution of α-solids G/T Multi-Triangulation Adaptation with Stellar and Handle Operators Luiz Velho October,

13 Stochastic Sampling/Reconstruction Poisson-Disk Distribution Random Uniform Sampling Irregular Pattern Points r-distant, at least r Reconstruction without Aliasing Combined Low-Pass Filter * Natural Scale / Resolution Luiz Velho October,

14 Sampling Solid Regions * Poisson-Disk Sampling is only Defined for Restriction to Must take into account does not follow geometry Luiz Velho October,

15 Two-Steps Algorithm (1) 1. Sample the Boundary 2. Sample the Interior minus r-tubular Neighborhood r * Stratified Sampling - Feature Sensitive Implementation - Dart Throwing - Quad-Tree Acceleration Luiz Velho October,

16 Multiresolution Sampling Spaces Nested Poisson-Disk Scale Spaces Disk Radius Multiresolution Hierarchy P 1 P P 0-1 Luiz Velho October,

17 Multiresolution Solid Sampling Apply Algorithm (1) for Tagged Samples: Boundary / Interior Resolution Level S 4 S 2 S 1 Luiz Velho October,

18 Structuring and Reconstruction Piecewise-Linear Approximation α-solids (α-shape + Regularization) * Subset of Delaunay Triangulation circumradius < α Luiz Velho October,

19 Looking for the Right Operators Operations on Simplicial Complexes Building Change Resolution Change Topology Types of Operators Topological Handlebody Theory Geometric Stellar Theory Luiz Velho October,

20 Handle Operators Change Topology Connected Components create Boundary destroy glue unglue Luiz Velho October,

21 Stellar Operators Change Combinatorial Structure Resolution split Structure weld flip Luiz Velho October,

22 Integrated Framework Combinatorial Manifold Operators Handle Stellar Properties Atomic Minimal Set Consistent (by construction) * Effective Abstractions Luiz Velho October,

23 Reconstruction from α-sampling Advancing Front Triangulation Ball-Pivoting Algorithm (2) while (points to process) while (e=candidate edge) p = ball_pivot e p create σ p glue σ p if (t=find seed) create σ t * Obs: Roll α-ball in Luiz Velho October,

24 Topological Guarantees Theorem (Amenta, 1999): If P r is an r-sample for r < k.lfs(s), then the crust is homeomorphic to S. * Applies to α-solids, (α = r) medial axis LFS(q) q r-sample r-solid Luiz Velho October,

25 Quality Guarantees Theorem (Medeiros et al, 2007): Let P α be an α-sampling of S, then the aspect ratio of C α (P α ) is bounded by. Worst case: a = b = r = α r Luiz Velho October,

26 Sampling Independence Different α-samplings Aspect Ratio Distribution Luiz Velho October,

27 α-filtration Defining the Multiresolution of Solids Ball-Pivoting at Each Level How to Move Gradually Between Levels? Change radius: 2 j < α < 2 j+1 Apply Stellar and Handle Operators * g(α) defines an Order for Samples Construction Strategies: Refinement Simplification Luiz Velho October,

28 Top-Down Construction Point Insertion Algorithm (3) while (samples to insert) if (inside) split σ p else create σ p glue σ p while (not Delaunay) switch (condition) case 1: flip e case 2: destroy σ t Luiz Velho October,

29 Bottom-Up Construction Point Removal Algorithm (4) while (samples to remove) while (valence 3) flip e if (inside) weld σ t if (boundary) create σ p weld σ p Luiz Velho October,

30 Coarse Grain Global Resolution Levels Luiz Velho October,

31 Intra Levels Fine Grain Local Luiz Velho October,

32 Properties Theorem (Uniqueness) : Given S α and g, the α-filtration generates a unique family of triangulations T α Theorem (Symmetry) : In α-filtration, the refinement sequence is the inverse of the simplification sequence insert p 1 split t 1 flip e 1 insert p 2 create t 2 glue t 2... insert p n split t n weld t n remov p n... unglue t 2 destroy t 2 remov p 2 flip e 1 weld t 1 remov p 1 Luiz Velho October,

33 G/T Multi-Triangulation Operators (stellar / handle) make only local changes (geometry / topology) Changes are weakly interdependent! Multi-Triangulation Structure Encodes Hierarchy Dependencies Representation for All Possible Meshes! Luiz Velho October,

34 Lattice Geometric Hierarchy Interior Decomposition Luiz Velho October,

35 Graph Topological Hierarchy Boundaries Luiz Velho October,

36 Geometric Dependencies Incident Faces at Same Level weld flip Topological All Edges must be Interior weld flip Luiz Velho October,

37 Mesh (half-edge) Data Structure h0 h1 h1 h3 h2 v h edge face vertex Multi-triangulation face-vertex edge-edge int-bd edge Luiz Velho October,

38 Adaptation Spatially Variant Function Domains Geometric Boundary (Shape) Interior (Attributes) Structure Components (Topology) Combinatory (Quality) Luiz Velho October,

39 Covering Elements Partition the Mesh in Regions simp.face ref.vertex int.edge int.edge int.edge bd.edge Priority Queues Sort Regions based on Adaptation Function refinement simp-faces int-edges int-edges simplification ref-vertices int-edges bd-edges Luiz Velho October,

40 Maintaining the Queues Algorithm (5) while (top_val > target) pop queue apply transition update elements Same Algorithm for All Tasks! Refinement / Simplification / Improvement Transition includes Dependencies Update re-evaluates Changes Luiz Velho October,

41 Algorithm (6) Dynamic Adaptation T 0 = base_triangulation initialize priority_queues while (adapting) read parameters evaluate f on T k simplify T k refine T k improve T k Algorithm (5) * Obs: Conservative! Luiz Velho October,

42 Adaptation Criteria Single Criteria (defaults for others) Multiple Criteria (must solve conflicts) Applications Examples Attribute Resolution Topology Granularity Geometry Detail Luiz Velho October,

43 Area of Triangles Attribute Example Luiz Velho October,

44 Topological Example Size of Holes Luiz Velho October,

45 Future Work Applications Extend to 3D Final Remarks Conclusions Theoretical Results Computationally Efficient Effective in Practice Luiz Velho October,

46 Thanks! Luiz Velho October,