improved cluster analysis

search radius cluster algorithm

• fastest implimentation is hist3 with binsize = search radius

• can easily require multiple molecules in the search radius by filtering:

  M=hist3([x,y],[Xmin:radius:Xmax,Ymin:radius:Ymax]);
  so if we want 15 nm search radius, just make 15nm steps

  M>3 require at least 3 molecules in each 15 nm “search radius” (i.e. bin).

• However this is going to crash the memory to generalize straight to 3D with same search radius — (matrix 4000 x 4000 is fine, 4000^3 is not).

• so we use the above approach to 2D cluster the raw image data to ID initial clusters for investigation, and 3D histogram these guys.

• matlabs morphological algorithms like bwareaopen, regionprops etc work in 3D!

wrote: hist4

simple extension of hist3 cluster algorithm. A beautiful inroad to all of matlabs image processing options, including regionprops, watershed, imfilter. Note though matlab lacks fspecial 3D there is a quality implementation here: fspecial3