mia-3dsegment-local-cmeans {-i io} [
options
...]
This program runs the segmentation of a 3D image by applying a localized c-means approach that helps to overcome intensity inhomogeneities in the image. The approach evaluates a global c-means clustering, and then separates the image into overlapping regions where more c-means iterations are run only including the locally present classes, i.e. the classes that relatively contain more pixels than a given threshold. This program implements algorithm described in:
image to be segmented. For supported file types see Plugin type: 3dimage/io
class label image based on merging local labels. For supported file types see Plugin type: 3dimage/io
class label image based on global segmentation. For supported file types see Plugin type: 3dimage/io
class probability image file, filetype must support floating point multi-frame images. For supported file types see Plugin type: 3dimage/io
verbosity of output, print messages of given level and higher priorities. Supported priorities starting at lowest level are:
| debug: | Debug output |
| message: | Normal messages |
| info: | Low level messages |
| error: | Report errors |
| fatal: | Report only fatal errors |
| trace: | Function call trace |
| warning: | Warnings |
| fail: | Report test failures |
print copyright information
print this help
print a short help
print the version number and exit
Spacing of the grid used to modulate the intensity inhomogeneities
c-means initializer. For supported plug-ins see Plugin type: 1d/cmeans
c-means breaking condition for update tolerance
threshhold to ignore classes when initializing the local cmeans from the global one.
Don't take background probablities into account when desiding whether classes are to be ignored
for values <= 0.5: create segmentation based on highest class probability, labels start at 0. For values >0.5: create labels only for voxels with a class probability higher than the given value, labels start at 1 and voxels without an according class probability are set to 0; this output is suitable for the seeded watershed filter.
Maxiumum number of threads to use for processing,This number should be lower or equal to the number of logical processor cores in the machine. (-1: automatic estimation).
Run the segmentation on image test.png using three classes, local regions of 40 pixels (grid width 20 pixels), and a class ignore threshold of 0.01.
mia-3dsegment-local-cmeans -i test.png -o label.png -n 3 -g 20 -t 0.01
Gert Wollny