Hi Roger ... As a compromise, if you are normally scanning in lines parallel to the x-axis. Then you would only need to store the current and previous two
Message 1 of 9
, Nov 1, 2008
On Thu, 30 Oct 2008, vortexswirling wrote:
> Since your algorithm considers north-south-east-west values. Instead of
> scanning the sphere one point at a time line by line, maybe scanning
> diagonal plus sign like patches would allow this to be done at run time
> with no need to store more than two points. From one scan patch of C
> (center) NSEW are also done. Then the next patch goes up one and over
> one, the previous N becomes W and the previous E becomes S, so those
> two points don't need to be recalculated.
> I think in the end each point still needs to be calculated twice, so
> complexity goes to the time domain.
As a compromise, if you are normally scanning in lines
parallel to the x-axis. Then you would only need to store
the current and previous two lines of results to be able to
check the previous line for its neighbours' z-values.
If you also discard the points whose neighbours are within
R/sqrt(2) then you will be storing very few points.
> What if verts of the geom could be indexed on x and y? Just thinking
> out loud...
This is what I had in mind when I suggested a map of the x,y
coordinates to the two z-values.