[SoC] Image Segmentation - GRASS - GSoC report #7
eric.momsen at gmail.com
Fri Jul 6 18:27:57 PDT 2012
Accomplished both planned tasks:
1. An optional size threshold, to force a final merge for very small
segments (even when not similar) makes very nice looking segments. (It
will be up to the user to decide if more cohesive segments are worth the
slight increase in heterogeneity.)
2. Speed increase. I was using a (fast) processing flag for the entire
raster, but only needed to keep track of some local pixels. It was an
O(area^2) speed cost, that was used repeated... so changing to a "slow"
tree structure for just the local information should a huge speed
improvement. (Thanks Markus for noticing this!) This change was just
implemented, still need to do some performance testing.
Step back and evaluate what features/options are most important to
implement in the last month of coding.
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the SoC