logo
  • Home   
  • Research Group Profile   
  • Applications   
  • Services   
  • Contact Us   
  • Multiscale Medial Axis Map

    • Let U be a non-empty closed subset of the Euclidean space Rn. The medial axis M(U) of U is the set of points x in the complement of U that have at least two closest points in U. If U is a bounded open set, then M(U) is the set of points in U with at least two closest points in the boundary of U. Bounded open sets are next referred to as shape. The medial axis provides a compact representation of the shape, and has been used in a number of applications. It is, however, well known that such geometric structure is unstable, for instance, with respect to the Hausdorff distance. A small change in shape can cause a relatively large change in the corresponding medial axis. Furthermore, it must be observed that in many aplications the boundary of a shape might also not well be defined but given through sampling. In this case, therefore, the challenge is to compute an approximation of the medial axis which catches the same topological type as of the original shape and is more stable under smmall perturbations of the shape.
      The multiscale medial axis map Ml(U), with l, provides a stable continuous representation of the medial axis with respect to Hausdorff measure and can remove undesirable parts of the medial axis by simple thresholding. The multiscale medial axis map Ml(U)=Ml(x;U) is a scalar function with support in a neighborhood of the medial axis M(U). The values of Ml(x;U) and its curvature depend on the strength of the boundary and on the parameter l which also controls the support width of Ml(U).
  • EXAMPLES:
    • On this page, we display some numerical simulations for finding the multiscale medial axis map of 2D shapes. For applications to 3D shapes, please visit the link 3D Multiscale Medial Axis.
    • i) Binary image and branching hierarchy
    • ii) Dense uniforn sampling of the boundary
    • iii) Coarse non uniforn sampling of the boundary
    • iv) Coarse non uniforn random sampling of the boundary
  • i) Binary image and branching hierarchy
  • chinese symbol for an
    (a)
    chinese symbol for an
    (b)
    • (a) Binary image (jpeg, with size 786x785) representing the chinese symbol 'an' together with the support of the multiscale medial axis map Ml(U)=Ml(x;U). In this picture, Ml(U) has been obtained for l=0.1 and it=1, where it is the number of iterations for the scheme used to compute Ml(U).
    • (b) Suplevel set of Ml(U) corresponding to a level value equal to 30, l=0.1 and it =1. That is to say, the picture displays the set {x: Ml(x;U)>30} overlapped to the input binary image.
  • chinese symbol for an
    (a)
    chinese symbol for an
    (b)
    • (a) Support of the multiscale medial axis map Ml(U)=Ml(x;U) for l=4 and it=1.
    • (b) Difference between the input binary image and the reconstructed image. The difference amounts to 36 pixels out of 69315 composing the binary image.
    • Variation of the number of pixels with threshold
      Diagram displaying for different values of the parameter l, the variation with the threshold of the number of pixels belonging to the difference between the input image and the reconstructed one from Ml(U) computed assuming it=1. The total number of pixels composing the binary image is 69315.
  • ii) Dense uniforn sampling of the boundary
  • Discrete set of points on ellipse boundary
    (a)
    Support of the MMAM
    (b)
    • (a) Discrete set U of points from a dense uniform sampling of the boundary of an ellipse.
    • (b) Support of the multiscale medial axis map Ml(U)=Ml(x;U) for l=0.1 and it=1.
  • Discrete set of points on ellipse boundary
    (a)
    Support of the MMAM
    (b)
    • (a) Suplevel set of Ml(U) corresponding to a level value equal to 32, l=0.1 and it =1.
    • (b) Reverse distance transform corresponding to the suplevel set of Ml(U) displayed in (a).
  • iii) Coarse non uniforn sampling of the boundary
  • Discrete set of points on ellipse boundary
    (a)
    Support of the MMAM
    (b)
    • (a) Discrete set U of points from a coarse non uniform sampling of the boundary of an ellipse.
    • (b) Support of the multiscale medial axis map Ml(U)=Ml(x;U) for l=0.1 and it=1.
  • Discrete set of points on ellipse boundary
    (a)
    Support of the MMAM
    (b)
    • (a) Suplevel set of Ml(U) corresponding to a level value equal to 48, l=0.1 and it =1.
    • (b) Reverse distance transform corresponding to the suplevel set of Ml(U) displayed in (a).
  • iv) Coarse non uniforn random sampling of the boundary
  • Discrete set of points on ellipse boundary
    (a)
    Support of the MMAM
    (b)
    • (a) Discrete set U of points from a coarse non uniform random sampling of the boundary of an ellipse.
    • (b) Support of the multiscale medial axis map Ml(U)=Ml(x;U) for l=0.1 and it=1.
  • Discrete set of points on ellipse boundary
    (a)
    Support of the MMAM
    (b)
    • (a) Suplevel set of Ml(U) corresponding to a level value equal to 56, l=0.1 and it =1.
    • (b) Reverse distance transform corresponding to the suplevel set of Ml(U) displayed in (a).

Home | Profile | Applications | Terms of Service | Contact Us
© 2011 KEA All Rights Reserved