Ouch, XML meshes :).
I actually had a good look at this, my first answer was pretty lazy. you can write better (as posted above), and it's not that complicated, I wouldn't fork out $40 for a journal article over this. Here's a pseudo-code solution which should work for you.
Note: When I say table, I mean 'lookup table'.
Assume each triangle is numbered and composed of vertices, v1, v2, v3, which are uniquely numbered and can be compared using the < operator (so we can obtain unique key-combinations).
We need two look-up tables:
- An edge->(triangle list) table called edge_triangles.
- A triangle->(edges list) table called triangle_edges.
A table which tells us which triangles use a given edge, and another which tells us which edges a a given triangle is comprised of. We build these lists as follows:
for t = next triangle
// Determine the ordering of vertices.
min_vertex = min(t.v1, t.v2, t.v3);
mid_vertex = median(p.v1, t.v2, t.v3);
max_vertex = max(t.v1, t.v2, t.v3);
// Register which edges this triangle uses.
edge_triangles[min_vertex][mid_vertex].append(t);
edge_triangles[mid_vertex][max_vertex].append(t);
edge_triangles[min_vertex][max_vertex].append(t);
// Set the edges that make up this triangle.
triangle_edges[t].append({min_vertex, mid_vertex});
triangle_edges[t].append({mid_vertex, max_vertex});
triangle_edges[t].append({min_vertex, max_vertex});
for next t
Using these lists, we can take the edges in a given triangle, use these as the key into the edge table, and see which polygons share that edge. Thus, the adjacent triangles. So for a triangle t we could do the following:
adjacent = edge_faces[face_edges[t][0]];
which is pseudo-code for 'adjacent is equal to the list of triangles that share the 0th edge of the triangle t', where 0th is just the first.
We use min, median and max to make sure that that we don't have different entries for identical edges: for example {v1, v2} and {v2, v1}, where v1 and v2 are two vertices. We could actually ignore this and add a 'compact' step, where we concatenate lists which correspond to different entries in our edge list but actually correspond to the same edge.
One other possible problem with this is if you have two edges which are coincident but don't share common vertices. In this case, you could reduce either edge to a parametric equation, compare them for coincidence, and form a lookup table which tells you, for a given edge, which edges are coincident, so map:
- edge->(edges list) table called edge_coincident_edges.
We use yet another look-up table because we can't concatenate the edge->faces table. Cosider if edges e1 and e2 are adjacent, e2 and e3 are, but e1 and e3 aren't. if we concatenated the e1, e2 and e3 entries in the edge->face list, you'd end up with some wildly incorrect data. This a probably bit more than you want to do, but it's the problem I had to solve this morning :).
In the case where each edge can only correspond to at most 2 triangles, we can do away with the 'list' in the traditional sense that we can append, and use a fixed-size array of size 2. This will reduce your memory overhead and improve memory efficiency. so our edge table would be more akin to:
- An edge->(first triangle, second triangle) table called edge_triangles.
Anyway, the basic algorithm is extensible to any number of polygons with any number of edges (not necessarily the same between all polygons), and is O(N) time with respect to the number of triangles (or polygons in the general case). Space complexity is O(E + N) Where E is edges and N is the number of polygons. The look-up time should be close to O(1), assuming you have good hashing algorithms.
