Original DuplicateDots based on course graining and overlapping pixels not very precise results:
Here green circles mark the red dots and yellow pluses mark the blue dots. The two colors are confocal stacks separated by ~400nm. As you can see from this image, a reasonable fraction of dots are still visible in both layers.
Green squares identify blue dots which are determined by the original course graining code DuplicateDots.m detection algorithm. The course graining looses substantial centroid information.
The lower pair shows a correctly identified duplicate (labeled as overlapping), but some duplicates are still missed, such as the closely aligned pair labeled non-overlapping.
Now open blue circles mark all the blue dot centroids located and yellow pluses mark the red dot centroids (flipped from above, I think this color choice more natural). Filled in circles are near the hollow circles if the dot is deemed to be unique.
A strel filter of radius 2 is used to determine overlap. Here’s a zoomed out view:
Much more reliable detection. I’ve also improved the dotfinder parameters to do better centroid detection / dot separation than achieved with previous posts.
Next need to test if this method can allow lower threshold choice in dot detection and weak dots to be eliminated by accurate assignment to other layers.