This seems similar to the pyramid model of Graham, Joshi, Pizlo (2000).
Modeling how (they think) human vision works, they made a heuristic based on clustering and top-down refinement. The map is clustered into a small handful of groups, and these groups are placed in a shortest tour. Each cluster is then clustered in turn into another handful of sub-groups, and these sub-groups are inserted into the tour at the position of their parent.
The pyramid algorithm is highly parallel with, IIRC, the same time characteristics as the space-filling curve algorithm.
What made the connection in my mind was that the simplest implementation of the pyramid algorithm is to cluster by simply dividing the map into quadrants; when you do so, you end up with the same space-filling curve they use at the top of the page.
Modeling how (they think) human vision works, they made a heuristic based on clustering and top-down refinement. The map is clustered into a small handful of groups, and these groups are placed in a shortest tour. Each cluster is then clustered in turn into another handful of sub-groups, and these sub-groups are inserted into the tour at the position of their parent.
The pyramid algorithm is highly parallel with, IIRC, the same time characteristics as the space-filling curve algorithm.
What made the connection in my mind was that the simplest implementation of the pyramid algorithm is to cluster by simply dividing the map into quadrants; when you do so, you end up with the same space-filling curve they use at the top of the page.