Some notes on the Sugihara&Iri 1994 topology-based algorithm for incremental construction of the voronoi diagram for point sites.
(A) We start with initial VD for point sites that have already been inserted. We want to insert a new site, shown as a yellow sphere.
(B) We then find the VD face to which the new vertex belongs. Among the vertices that bound this face we search for a seed-vertex, shown in pink. This is the vertex that is closest to the new generator.
(C) From the seed vertex, we search for more vertices that should be deleted(red). The vertices should form a tree (a connected, acyclic graph). Vertices that are closer to the new site than to any other site should be deleted. However due to floating-point errors it's not possible to blindly rely on an inCircle() predicate for finding the delete-tree. It's necessary to also enforce the correct topology, namely (i) the delete-vertices should form a tree, and (ii) for any incident face, the delete-vertices are connected. This ensures that no old face of the graph is deleted or split in two.
(D) Identify the edges to be deleted (red), and edges on which we need to generate new vertices (green).
(E) Generate new vertices on the green edges, and connect all the new vertices by new edges. These new edges form a new face corresponding to the newly inserted site.
(F) Remove the red/delete-vertices and edges. We are left with what we want: the VD of all the old sites and the new site.
Another picture from the original 1994 paper. The new site is not shown, but by some process we have found the delete-tree (u1,u2,u3,u4) shown as solid dots. New vertices are then generated on the "in-out" edges (u2-u10, u2-u5, u3-u6, u3-u7, u4-u8, u4-u9), shown as unfilled dots. The unfilled dots are connected with new edges (dashed lines) that form the new facet (shaded area). Image borrowed from: Sugihara&Iri 1994.
Here is an animation of the algorithm at work when inserting about 100 random points:
Thanks for taking the time to explain the basics of the algorithm!
The visualization is really well done and helps (me) a lot.
Personally I highly appreciate the fact that you don't just focus on the funny part (programming), but also contribute to the understanding of the algorithms!