My guesses at solutions to the 3-cell marking problem

In this post, I would like to ponder on the possible solutions to the problem I presented in my previous post. Just as a recap, the problem is: Given a tetrahedral mesh and a surface mesh(the region of interest or ROI), we need to figure out a way to mark each tetrahedron of the given tetrahedralization either as inside or outside ROI. To avoid any confusion I should point out that the surface mesh representing the ROI is defined over vertices of the given tetrahedralization. Further, the title of the post refers to tetrahedrons as 3-cells similarly you can refer triangles as 2-cells following the terminology used here. As I have discussed here, I will first attempt to solve this problem in two dimensional domain, where my task simplifies as: Given a triangle mesh and a polygon boundary as input, mark all triangles of the mesh as either inside or outside the boundary.

An obvious approach to this problem is the extension of the existing solutions to Point in Polygon problem. Point in polygon problem is concerned with marking an arbitrary query point as either inside or outside the input polygon.

Common approach called the ray casting algorithm involves casting a ray in random direction from the query point and compute number of intersections it had with the boundary of the polygon. If the number of intersections are even it implies that the point is inside else the point is marked as outside the polygon boundary. Degenerate condition for this algorithm is when the point lies very near(or onto) to the boundary of the polygon, in which case the marking based on number of intersections fails in general.

However, in our case the problem is simplified by the following facts:

  1. Point will not be arbitrarily positioned but a representative of the query triangle say its barycenter.
  2. If the polygonal boundary is convex we do not have to deal with this problem at all because all triangles will be inside.

So, subsequent discussion has significance only for concave boundaries. An algorithm based on ray casting approach for marking triangles can be like this:

1 Add all triangles of the mesh in a queue 'Q'
  1.1 For each triangle 'T' in 'Q':
    1.1.1 Compute Barycenter 'b'
    1.1.2 Generate a ray 'r' in random direction emanating from 'b'
    1.1.3 Compute intersections with each segment of the polygonal boundary
    1.1.4 Count the number of unique intersections, i.e., discard identical intersection points.
    1.1.5 If the number of intersections are 'even', mark 'T' as 'outside', else mark 'T' as 'inside'  
2 Repeat 1 until 'Q' is not Null 

Currently, this seems to solve the marking problem in two-dimensional domain but I am keen to have your comments on this. Assuming this approach to be correct, I come back to my original problem of marking 3-cells as inside or outside the given ROI. I think it only amounts to extending above presented approach to 3D domain by replacing triangles T with tetrahedrons and the ROI becomes a polyhedron instead of a polygon.

Specifically, a 3D version of the above algorithm may look like:

1 Add all tetrahedrons of the tetrahedralization in a queue 'Q'
  1.1 For each tetrahedron 'T' in 'Q':
    1.1.1 Compute Barycenter 'b'
    1.1.2 Generate a ray 'r' in random direction emanating from 'b'
    1.1.3 Compute intersections with each polygon(in our case they are triangles) of the polygonal boundary
    1.1.4 Count the number of unique intersections, i.e., discard identical intersection points.
    1.1.5 If the number of intersections are 'even', mark 'T' as 'outside', else mark 'T' as 'inside'  
2 Repeat 1 until 'Q' is not Null 

Rough tracing on paper on some toy input examples seems to suggest this approach should work in case of 3D domain as well. But I will be implementing it now. I will report results in the posts to follow.

Written on 12 Mar 2015