``Cue-based_Relaxation_Labeling'' -- segment an image using relaxation labeling

purpose:

Uses relaxation labeling to segment an image into 2-8 distinct regions. The algorithms used to compute the initial probability vectors, perform the updating, and compute a final segmentation can be found in [7] and [8]. Briefly, using the naming as is found in [8], the strategies used for this version of relaxation labeling are: initial probabilities--- By-slice, vector updating--- fast general using support-reject compatibility coefficients. For final classification, each voxel gets the label corresponding to the largest vector component in that voxel's probability vector.

input:

An 8-bit volume called imagename, loaded into ROI using ROI's load volume command in the GO menu.

Another two 8-bit volumes, called imagename.b and imagename.c, which contain cue information that has been generated using the INTERSEG package [26].

output:

An 8-bit volume in segmentation volume format (refer to the demask region function, p. ) containing the segmented regions.

parameters:

Default parameters menu is

gray-scale input assumed from vol #0
* number of regions to segment = 2
* number of updating iterations = 10
* use exclusionary cues = no
* final segmentation threshold = 0.000
*

If the final segmentation threshold is set to 0.0, then the maximum vector component for each voxel is assumed to represent that voxel's appropriate labeling. If a threshold is set, then the maximum vector component value for a voxel must be greater than the threshold value for that voxel to be labeled as part of a region.

comments:

 

1. This ``turnkey'' version of the relaxation labeling function is described in detail in [8].