SCATTERED DATA VISUALIZATION Yingcai Xiao Scattered Data: sample points distributed unevenly and non-uniformly throughout the volume of interest. Example Data: chemical leakage at a tank-farm. Method of Approach : Interpolation-based Two-step Approach (Foley & Lane, 1990) Rendering Intermediate Grid Sparse Data Rendered Interpolation Grid-Based Volume Modeling Interpolation Methods (Nielson, 1993) Global: all sample points are used to interpolated a grid value. Local: only nearby sample points are used to interpolated a grid value. Exact: the interpolation function can exactly reproduce the data values on the sample points. Problems: Xiao etc. 1996 Interpolation Methods Example: 1D Global and Exact Interpolation Methods Example: 1D Global and Exact 2 f (x) =å bi x i=1 Defining a Global Exact Interpolant (Foley & Lane, 1990; Nielson, 1993) N sample points: (xi,yi,zi,vi) for i = 1,2,..n One interpolation function, e.g., Thin-plate spline, n f ( x, y, z ) = bi d log( di ) + c1 + c2 x + c3 y + c4 z i=1 2 i di is the distance between sample point i and the point to be interpolated p(x,y,z). di = ((x-xi)2+(y-yi )2+(z-zi )2)1/2 bi,c1,c2,c3,c4 are n+4 constants to be solved by enforcing the following conditions: f (xi,yi,zi) = vi for i = 1,2,..n Global Exact Interpolation Functions (Foley & Lane, 1990; Nielson, 1993) Thin-plate spline n f ( x, y, z ) = bi d log( di ) + c1 + c2 x + c3 y + c4 z 2 i i=1 n Volume Spline f ( x, y, z ) = bi d 3 + c1 + c2 x + c3 y + c4 z n f ( x, y, z ) = b d i i =1 Multiquadric Shepard 3 i + c1 + c2 x + c3 y c4iz, i =1 n Thin-plate Spline f ( x, y, z ) = bi d log( di ) + c1 + c2 x + c3 y + c4 z i =1 2 i Volume Spline n f ( x, y, z ) = b d i i =1 3 i + c1 + c2 x + c3 y + c4 z Shepard method f ( x, y, z ) = n 1 d i vi i =1 n 1 di i = 1 Deficiencies of the Interpolation-based Two-step Approach (Xiao et. Al., 1996) l Misinterpretation (Negative Concentration) l Ambiguity in Selecting Interpolation Methods l Inconsistent Interpolations in Modeling and Rendering l Visualizing Secondary Data Instead of the Original Data l No Error Estimation l Unable to Add Known Information l Not Efficient Three Dilemmas and Three Constraints (Xiao & Woodbury, 1999) l Zero-value dilemma l Negative-value dilemma l Correctness dilemma lPoint Constraint l Value Constraint l Local Constraint Point Constraint v sample points constraining points d v extrapolated values sample points d Value Constraint v v min, if f ( x , y , z ) < vmin, f ( x , y , z ), v max, if f ( x , y , z ) > vmax. p6 p1 Local Constraint p2 p7 p3 p4 p5 p8 Conclusions • Two-step approach faces three dilemmas. • Constrained interpolations can alleviate the dilemmas. • The problems are far from being solved. Data modeling is import to data visualization, just as geometry modeling is important to geometry visualization. Conclusions To visualize scattered data, we are challenged to find modeling techniques that l preserve input data values; l produce meaningful output values; l provide error estimations; l accept additional constraints; l reduce the requirement on the sampling intensity. A FINITE ELEMENT BASED APPROACH XIAO & ZIEBARTH, 2000 The Finite Element Based Approach (1) Tessellation (2) Computation (3) Rendering The Finite Element Based Approach Rendering Tessellation Computation Rendered Volume Node Values Sparse Data Volume Element Network Triangulation FEM Element-Based Tessellation lThree-Dimensional Triangulation: Tetrahedronization lDelaunay Triangulation: Sphere Criterion in put samp le p oint s input sample point s discont inuit y point s refinem ent point s discont inuit y surface Data Points disco nt inuit y point s disco nt inuit y surface refin em ent po int s t rian gulat ed n et work Triangulation Element Network The Double Layer Technique in put samp le p oint s in put samp le p oint s disco nt inuit y point s disco nt inuit y surface refin em ent po int s t rian gulat ed n et work Physical Discontinuity disco nt inuit y point s do uble layers refin em ent po int s t rian gulat ed n et work Logical Discontinuity The Finite Element Method (1) Problem Definition: l Boundary Value Problem L f Governing equation: p on S l Boundary Condition: (2) Element Definition: l Shape: Tetrahedron 4 l Order: Basis Function ( x, y, z) N ej ( x, y, z) ej e j 1 The Finite Element Method (3) System Formulation l l Ritz Method Galerkin's method [K]{f } = {C}. (4) Sparse Sample Data (5) System Solution l l Gaussian Elimination Householder's Method F ( ) 21 L , 21 , f 21 f , r L f {} = {i, i=1,2,...,n}T k p( k ) Rendering : Modifying Conventional Methods (1) Hexahedron => Tetrahedron (2) (ijk) Indexing => Neighbor-to-Neighbor Traversal Advantages of the Finite Element Based Approach (1) Meaningful Results Z Ground Surface 2000 1000 1000 0 Y 1000 X A Pollution Problem Exact Grid-based FEM-based Advantages of the Finite Element Based Approach (2) Complicate Geometry: Non-Gridable Volumes Advantages of the Finite Element Based Approach (3) Discontinuity: Internal Discontinuity Surface Advantages of the Finite Element Based Approach (3) Discontinuity: Discontinuous Regions Advantages of the Finite Element Based Approach (4) Error Estimation and Iterative Refinement E e 21 | '' | h 2 h 2 E lim /| '' | 4 3 2 1 0 0 500 1000 1500 2000 Z h Error 1.0 1.0 0.5 0.25 0.25 0.0625 Advantages of the Finite Element Based Approach (5) Efficient Add One Point => Add O(1) Tetrahedrons O(n2) Times More Efficient Than Grid-Based Approaches. Advantages of the Finite Element Based Approach (6) No Whittaker-Shannon Sampling Rate Interpolation Problem ==> Boundary Value Problem (7) No Ambiguity in Selecting Modeling Methods Advantages of the Finite Element Based Approach (8) Honoring Original Sample Data Advantages of the Finite Element Based Approach (9) Flexible, Fast and Interactive Modification of an Existing Sample Point Advantages of the Finite Element Based Approach (9) Flexible, Fast and Interactive Addition of a New Sample Point Advantages of the Finite Element Based Approach (10) Consistent Basis Function 4 ( x , y , z ) N ej ( x , y , z ) ie e j 1 1 N ej ( x j , y j , z j ) ij 0 i j i j Future Work (1) Other Types of Problems: Initial Value Problems (2) Other Types of Elements: Polyhedrons (3) Higher-Order Elements: P-Version (4) Automated Tessellation: Densification (5) Thinning (6) Curved Discontinuity Surfaces (7) Delaunay Triangulation near Discontinuity Surfaces (8) Higher-Order Rendering Method (9) Fast Searching Algorithms (10) Technique Issues (e.g., Solving Sparse Matrices, ...) Summary The finite element based approach is a new framework for scattered data visualization. Many challenging problems can be solved easily within this framework. This approach revealed a promising direction and brought many interesting research topics into the field of sparse data volume visualization.