Links

  • Project Update for the October 16th deadline.
  • Project Update for the November 15th deadline.
  • Final report for the December 11th deadline.
  • Slides for the December 11th presentation.
  • Background Information
    M-Rep stands for medial representation. It is a 3D structure made up of a quad-mesh of medial atoms. (There are other medial representations, but this is the one I will focus on.) Each atom stores information about the implied 2D surface boundry as well as the figural direction of the object. Since the boundary of the object is actually implied from the 3D structure, operations related to the boundary surface like intersection computation could be difficult. The boundary surface is a grid of points obtained from the medial atoms connected with B-Spline patches.

    This M-Rep structure was created by the MIDAG group at UNC. They have a program called Pablo they use to construct and view M-Rep models. Some work has been done in Pablo to allow protrusions, objects with other objects attached to it. Also, for protrusions blending has been accomplished using a distance function, which performs slower then MIDAG would like.

    Goals
    My first goal is to implement the CSG (Constructive Solid Geometry) needed by the MIDAG group. This includes protrusion (Union) which has already been implemented by Shawn Liu, and indentation (subtraction) where there are constraints on the negative object. These constraints could include things like the medial atom mesh of the negative object must start on the boundry of the positive object and go in and stop before effecting the opposite boundry.

    My second goal is to implement blending for protrustion and indentation by changing the B-Spline patches near the intersection of the objects.

    My third goal is to implement the rest of the CSG operations like intersection and to not have any restrictions on subtraction, with blending, of course. This goal isn't necessary for MIDAG and will only be done if I have enough time. Also, if everything above turns out to be really easy there are addition problems to work on related to CSG using continuous M-Reps, where instead of a mesh of medial atoms you have a 2D manifold.

    I believe that the first goal will take the longest, because finding the intersection of the two surfaces will be used by both the CSG operations and the blending.

    The Time-Line for these goals follows in the order listed above:

    Related Work and Novel Aspects
    A bunch of work has been done on finding the intersections of two mesh surfaces, I have not yet looked for this material but am sure it exists. I don't believe any work has been done on this using M-Reps though. I am hoping that I can find a way to make finding the intersections more effecient for M-Reps by taking advantage of the 3D structure. For M-Reps the current way to blend two objects (protrusion only) together involves using a distance function. I hope that my way of editing the B-Spline patches is more effecient. I am not sure if this last part is novel to the industry, but it is novel to M-Reps and to myself.

    References

  • I talked with Pizer about this project and will also interact with Tom Fletch and Shawn Liu.
  • Most of my current knowledge of M-Reps came from the introduction in:
    Deformable M-Reps for 3D Medical Image Segmentation. Pizer et al. (2001).
  • I have also looked through the Pablo User Guide.