1 / 40

Global overview

Special course, 323.022 Level Set Methods and Dynamic Implicit Surfaces in Comp. Vision and Comp. Physics Arjan Kuijper 08.03.2007. Johann Radon Institute for Computational and Applied Mathematics (RICAM) Austrian Academy of Sciences Altenbergerstraße 56 A-4040 Linz, Austria

tale
Télécharger la présentation

Global overview

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Special course, 323.022Level Set Methods and Dynamic Implicit Surfaces in Comp. Vision and Comp. PhysicsArjan Kuijper08.03.2007 Johann Radon Institute for Computational and Applied Mathematics (RICAM) Austrian Academy of Sciences Altenbergerstraße 56A-4040 Linz, Austria arjan.kuijper@oeaw.ac.at

  2. Global overview • This course is an introduction to level set methods and dynamic implicit surfaces. • These are powerful techniques for analyzing and computing moving fronts in a variety of different settings. • Besides many examples of the utility of the methods to a diverse set of applications, also complete numerical analysis and recipes will be given, which will enable one to quickly apply the techniques to real problems.

  3. Material Level Set Methods and Dynamic Implicit Surfaces Stanley Osher & Ronald Fedkiw Series: Applied Mathematical Sciences, Vol. 153 Springer, 2003 ISBN-10: 0-387-95482-1 ISBN-13: 978-0-387-95482-0

  4. example • C:\Documents and Settings\arjan\Desktop\glass00.avi • http://www-sop.inria.fr/ariana/DEMOS/classif_levelset/demo.html

  5. Tentative schedule • Week 1: Implicit Surfaces: • Basic properties, level set geometry, signed distance functions. • Week 2-7: Level Set Methods: • Topics include the level set equation itself, Hamilton-Jacobi equations, motion of a surface normal to itself, re-initialization to a signed distance function, extrapolation in the normal direction, the particle level set method and the motion of co-dimension two (and higher) objects. • Week 8-9: Image Processing and Computer Vision: • The restoration of images degraded by noise and blur, image segmentation with active contours (snakes), and reconstruction of surfaces from unorganized data points. • Week 10-14: Computational Physics: • One phase compressible fluid dynamics, two-phase compressible flow involving possibly different equations of state, detonation and deflagration waves, and solid/fluid structure interaction; incompressible fluid dynamics, free surface flows, and fully two-phase incompressible flow, incompressible flames, coupling a compressible and incompressible fluid, heat flow and Stefan problems, all with applications to computer visualization.

  6. 08.03.2007 :Implicit Surfaces Implicit Functions Signed Distance Functions

  7. Implicit Functions - points • In one spatial dimension, suppose we divide the real line into three distinct pieces using the points x = -1 and x = 1. • W- = (-1, 1) is the inside portion of the domain. • W+ = (-,-1) (1, ) is the outside portion of the domain. • W= {-1, 1} is the border between the inside and the outside and is called the interface. • In one spatial dimension, the inside and outside regions are one-dimensional objects, while the interface is zero-dimensional. • More generally, in Rn, subdomains are n-dimensional, while the interface has dimension n - 1. We say that the interface has codimension one.

  8. Implicit Functions - points • In an explicit interface representation one explicitly writes down the points that belong to the interface • Alternatively, an implicit interface representation defines the interface as the isocontour of some function. • For example, the zero isocontour of f(x) = x2 - 1 is the set of all points where f(x) = 0. • Note that the implicit function f (x) is defined throughout the one-dimensional domain, while the isocontour defining the interface is one dimension lower. • More generally in Rn, the implicit function f(x) is defined on all x  n; its isocontour has dimension n - 1.

  9. Implicit Functions - points • there is nothing special about the zero isocontour. • The g(x)=1 isocontour ofg(x) = x2 defines the same interface. • The partial derivatives off are the same as the partial derivatives of g, since the scalar vanishesupon differentiation. • Thus, all implicit functions f(x) will be defined so that the f(x) = 0 isocontour represents the interface.

  10. Implicit Functions - curves • In two spatial dimensions, our lower-dimensional interface is a curve that separates R2 into separate subdomains with nonzero areas. • Here we are limiting our interface curves to those that are closed, so that they have clearly defined interior and exterior regions. • Example:f(x,y)= x2 + y2 – 1.

  11. Implicit Functions - curves • In two spatial dimensions, the explicit interface definition needs to specify all the points on a curve. • While in this case it is easy to do, it can be somewhat more difficult for general curves. • In general, one needs to parameterize the curve with a vector function x(s), where the parameter s is in [s0, sf]. • The condition that the curve be closed implies that x(s0) = x(sf). • A convenient way of approximating an explicit representation is to discretize the parameter s into a finite set of points s0 < · · · < si-1 < si < si+1 < · · · < sf, where the subintervals [si, si+1] are not necessarily of equal size.

  12. Implicit Functions - curves • The implicit representation can be stored with a discretization as well, except now one needs to discretize all of R2, which is impractical, since it is unbounded. • Instead, we discretize a bounded subdomain D of R2. • Within D, we choose a finite set of points (xi, yi) for i = 1, . . . ,N to discretely approximate the implicit function. • This illustrates a drawback of the implicit surface representation. • Instead of resolving a one-dimensional interval [s0, sf], one needs to resolve a two-dimensional region D.

  13. Implicit Functions - curves • More generally, in Rn, a discretization of an explicit representation needs to resolve only an (n - 1)-dimensional set, while a discretization of an implicit representation needs to resolve an n-dimensional set. • This can be avoided, in part, by placing all the points x very close to the interface, leaving the rest of D unresolved. • Since only the f(x) = 0 isocontour is important, only the points x near this isocontour are actually needed to accurately represent the interface. The rest of D is unimportant.

  14. Implicit Functions - curves • Neither the explicit nor the implicit discretization tells us where the interface is located. Instead, they both give information at sample locations. • In the explicit representation, we know the location of a finite set of points on the curve, but do not know the location of the remaining infinite set of points (on the curve). • In the implicit representation we know the values of the implicit function f at only a finite number of points. • Interpolation is needed to find exact locations.

  15. Implicit Functions - grid • The set of data points where the implicit function is defined is called a grid. • Cartesian grids: {(xi, yj) | 1  i m, 1  j  n}. x1 < · · · < xi-1 < xi < xi+1 < · · · < xm and y1 < · · · < yj-1 < yj < yj+1 < · · · < yn. • In a uniform Cartesian grid, all the subintervals are equal in size: x = xi+1 – xi, y = yj+1 - yj. • Choose x = y so that the approximation errors are the same in the x-direction as they are in the y-direction. • Cartesian grids imply the use of a rectangular domain D = [x1, xm]×[y1, yn].

  16. Implicit Functions - surfaces • Extend to 3D: f(x,y,z)=x2 + y2 + z2 – 1. • For complicated surfaces with no analytic representation, we again need to use a discretization. • In three spatial dimensions the explicit representation can be quite difficult to discretize. • One needs to choose a number of points on the two-dimensional surface and record their connectivity which is less straightforward. • Some of the most popular algorithms can produce surprisingly inaccurate surface representations, e.g., surfaces with holes. • Connectivity can change for dynamic implicit surfaces, i.e., surfaces that are moving around. Here, connectivity is not a one-time issue dealt with in constructing an explicit representation of the surface. • Instead, it must be resolved over and over again every time pieces of the surface merge together or pinch apart. • One of the nicest properties of implicit surfaces is that connectivity does not need to be determined for the discretization.

  17. Geometry Toolbox • Implicit interface representations include some very powerful geometric tools. • For example, since we have designated the f(x) = 0 isocontour as the interface, we can determine which side of the interface a point is on simply by looking at the localsign of f. • With an explicit representation of the interface it can be difficult to determine whether a point is inside or outside the interface.

  18. Geometry Toolbox • Numerical interpolation produces errors in the estimate of f. This can lead to erroneously designating inside points as outside points and vice versa. • These errors amount to perturbing (or moving) the interface away from its exact position. • If these interface perturbations are small, their effects may be minor, and a perturbed interface might be acceptable. • In fact, most numerical methods depend on the fact that the results are stable in the presence of small perturbations.

  19. Geometry Toolbox • Take numerical approximations with errors proportional to the size of a Cartesian mesh cell, x. • If the implicit function is smoothenough and well resolved by the grid, these estimates will be appropriate. • Otherwise, these errors might be rather large. Obviously, this means that we would like our implicit function to be as smooth as possible. • A signed distance function to represent the surface turns out to be a good choice, since steep and flat gradients as well as rapidly changing features are avoided as much as possible.

  20. Geometry Toolbox • The gradient of the implicit function is defined as • The gradient is perpendicular to the isocontours of f and points in the direction of increasing f. Therefore, if x is a point on the zero isocontour of f, i.e., a point on the interface, then f evaluated at x is a vector that points in the same direction as the local unit (outward) normal N to the interface. Thus, the unit (outward) normal for points on the interface is

  21. Geometry Toolbox • It will be useful to have as much information as possible representable on the higher-dimensional domain. • Embeds the normal in a function N defined on the entire domain that agrees with the normal for points on the interface.

  22. Geometry Toolbox • The normal is not defined in the origin where all partial derivatives vanish. • A simple technique for evaluating N at these points is just to pick an arbitrary direction for the normal as it is an isolated singularity. • The standard trick of adding a small  > 0 to the denominator can be a bad idea in general, since it produces a normal with | N |  1. • In fact, when the denominator is zero, so is the numerator.

  23. Geometry Toolbox • On our Cartesian grid, the derivatives need to be approximated, for example using finite difference techniques. • We can use a first-order accurate forward difference D+fa first-order accurate backward difference D-for a second-order accurate central difference D0f

  24. Geometry Toolbox • When all numerically calculated finite differences are identically zero, the denominator of N vanishes. • As in the analytic case, we can simply randomly choose a normal. Here, however, randomly choosing a normal is somewhat justified, since it is equivalent to randomly perturbing the values of on our Cartesian mesh by values near round-off error. • These small changes in the values of f are dominated by the local approximation errors in the finite difference formula for the derivatives.

  25. Geometry Toolbox • The mean curvature of the interface is defined as the divergence of the normal N = (n1, n2, n3),so that k > 0 for convex regions, k < 0 for concave regions, and k = 0 for a plane. • Using the definition of N:

  26. Geometry Toolbox • This gives k explicitly: • A second-order accurate finite difference formula for fxx, the second partial derivative of f in the x direction, is given by D+D-f = D-D+f : • A second order accurate finite difference formula for fxy is given by Dy0Dx0f, or equivalently, Dx0Dy0f. • Similar for fyx and fyy

  27. Geometry Toolbox • Again the origin is a removable singularity. • It seems nonsensical to be troubled by k as we approach the origin, since this is only a consequence of the embedding. • The smallest unit of measure on the Cartesian grid is the cell size Dx, it makes little sense to hope to resolve objects smaller than this. That is, it makes little sense to model circles (or spheres) with a radius smaller than Dx. Therefore, we limit the curvature so that • If a value of k is calculated outside this range, we replace it with the closest limiting case.

  28. Geometry Toolbox • The characteristicfunctionc- of the interior region W- is defined aswhere we arbitrarily include the boundary with the interior region. The characteristic function c+ of the exterior region W+ is defined similarly as

  29. Calculus Toolbox • These functions are functions of a multidimensional variable x. Thus define the one-dimensional Heaviside function • The volume integral of a function f over the interior region is defined asor

  30. Calculus Toolbox • The directional derivative of the Heaviside function H in the normal direction N is the Dirac delta functionwhich is a function of the multidimensional variable x. Note that this distribution is nonzero only on the interface where f = 0. We can rewrite it asIn one spatial dimension, the delta function is defined as the derivative of the one-dimensional Heaviside function giving

  31. Calculus Toolbox • The surface integral of a function f over the boundary is defined aswhere the region of integration is all of W, since it prunes out everything except the boundary automatically. The one-dimensional delta function can be used to rewrite this surface integral as • By embedding the volume and surface integrals in higher dimensions, the need for identifying inside, outside, or boundary regions is avoided. Instead, the integrals are taken over the entire region.

  32. Calculus Toolbox • The one dimensional delta function needs to be evaluated. Since d(f) = 0 almost everywhere, i.e., except on the lower-dimensional interface, which has measure zero, it seems unlikely that any standard numerical approximation based on sampling will give a good approximation to this integral. • Thus, we use a first-order accurate smeared-out approximation of d(f). • Define the smeared-out Heaviside functionwhere e is a tunable parameter that determines the size of the bandwidth of numerical smearing. A typically good value is e = 1.5Dx

  33. Calculus Toolbox • Then the delta function is defined as the derivative of the Heaviside function: • the smeared-out Heaviside and delta functions approach to the calculus of implicit functions leads to first order accurate methods. • Also when used with some integration method.

  34. Distance Functions • A distance function d(x) is defined asimplying that d(x) = 0 on the boundary. • Geometrically, d may be constructed as follows: • If x is on the boundary, then d(x) = 0. • Otherwise, for a given point x, find the point on the boundary set closest to x, and label this point xC. Then d(x) = |x - xC|. • The line segment from x to xC is the shortest path (steepest descent) from x to the interface. • Since d is Euclidean distance, |d| = 1. • Not unique, …but one of the triumphs of the level set method involves the ease with which these degenerate points are treated numerically.

  35. Signed Distance Functions • A signed distance function is an implicit function f with |f(x)| = d(x) for all x. • Thus: • f(x) = d(x) = 0 for all x  W , • f(x) = -d(x) for all x  W-, • f(x) = d(x) for all x  W+. • Furthermore, |f| = 1 • Only in a general sense, but: • Equations that are true in a general sense can be used in numerical approximations as long as they fail in a graceful way that does not cause an overall deterioration of the numerical method.

  36. example • For 1D:

  37. examples • For 2D: • And 3D: • Again: only singularities “far from the front”.

  38. Geometry and Calculus Toolboxes • Using |f| = 1: • And for the surface integral:

  39. Summary • Basics on implicit functions: • Implicit functions use an extra dimension • Consider the zero-isocontour • Define regions inside and outside using the Heaviside function and the boundary with a delta function • For discrete computations use a smeared version • Singularities in the normal vector can usually be ignored as they are ‘far away’ and removable. • The signed distance function give the Euclidian distance of a point to the boundary. • A very good choice for an implicit function

  40. Next week • Level Sets: • Motion in an Externally Generated Velocity Field • Convection • Upwind Differencing • Hamilton-Jacobi ENO • Hamilton-Jacobi WENO • TVD Runge-Kutta • Motion Involving Mean Curvature • Equation of Motion • Numerical Discretization • Convection-Diffusion Equations

More Related