Sparse blob detection/region-labeling in any number of dimensions for scientific language IDL.
Note: I have not used this code in YEARS, so I make no guarantee that it will work or even compile. I don't have access to IDL anymore. Still, if you find bugs I will try to fix them.
Performs connected-components region labeling, or "blob detection",
on binary image arrays with any number of dimensions.
Outputs a distinct integer label for each connected component.
The normal IDL function to do this is LABEL_REGION.
However, LABEL_REGION looks over the whole array.
This is very inefficient if you already know where the nonzero elements are located,
and there are relatively few of them.
This program consists of an IDL function, LABEL_REGION_SPARSE, and a few small helper functions in IDL.
The workhorse is an external C program, compiled to a DLL, that is called by LABEL_REGION_SPARSE
with IDL's MAKE_DLL and CALL_EXTERNAL capability.
###Calling Signature###
labels = label_region_sparse(array, indices, [/all_neighbors], [ndiff])
###Inputs###
ARRAY: the image data to scan for connected components.
INDICES: an array containing indices to the foreground elements in ARRAY,
e.g. where(ARRAY).
###Keywords###
NDIFF: sets the maximum number of dimensions in which a voxel's position
may be offset to be considered a neighbor. For example, in 3D:
NDIFF = 1 produces face connectivity.
NDIFF = 2 produces face and edge.
DIFF = 3 produces face, edge, and vertex (26-connectivity).
ALL_NEIGHBORS: switches on full connectivity: equivalent to setting NDIFF
equal to the number of dimensions of the array.
Behavior is exactly the same as IDL's LABEL_REGION.
If ALL_NEIGHBORS is set, it will override any setting of NDIFF.
If neither keyword is set, the default is NDIFF = 1.
###Output###
A vector of ULONGs, the same size as INDICES, containing the region label
for each element in INDICES. The regions are numbered 1..N with no gaps for an N-region image.