matsciseg
materials science image segmentation algorithms
Dependencies
-
Python 2.7 (go here for instructions to setup python 2.7 on machines with older versions of python)
-
OpenCV with Python bindings (tested with 2.3.1 and 2.4.3)
-
All packages under
requirements.txt, preferably installed into a virtualenv. A bug in some numpy/scipy package dependencies might mandate that numpy be installed manually -
(optional) wxWidgets (must be manually compiled and installed into virtualenv) to use the now-deprecated graphical interface
-
GCO compiled as a static library--patches for specific files, as well as a makefile that will build the shared library on linux are available under
gco_extra. -
Tested on Debian Stable (Squeeze), Fedora 15, Arch Linux
Setup
virtualenv --no-site-packages ~/.virtualenv
source ~/.virtualenv/bin/activate
# compile/install opencv and make it available to the virutalenv
pip install numpy==1.6.2 # fix circular dependency
pip install -r requirements.txt
make # fetches submodules, builds gco, and compiles gco wrappers
Use
A command line interface (matscicli.py) and a web interface
(webgui/manage.py runserver) are available.
The CLI relies upon different "recipes" in the recipes.py file,
which can be easily customized as needed. Its arguments are
documented upon invocation. The "label" files are plain text files
with rows (separated by newlines) of integers (separated by
whitespace) for each row of pixels in the corresponding image. Sample
invocations are shown in tests/tests.sh. The sample.sh script
will run the global algorithm on the data found in sample/ to
produce a file sample/0001-output.png, which is a good sanity check
that all is working as it should be.
The web interface requires volumes made of sequences of images each
with a label initialization (same format as above) converted into
individual numpy *.npz files. This can be done with the included
wrap_labels.py script. Placing these in
webgui/data/[data-name]/*.npz will allow the web interface to find
the data. To initialize the web interface, simply navigate to the
webgui folder, run
./manage runserver
and navigate to http://localhost:8000.
References
Techniques used in this package are described in:
-
Jarrell Waggoner, Youjie Zhou, Jeff Simmons, Ayman Salem, Marc De Graef, Song Wang. Interactive Grain Image Segmentation using Graph Cut Algorithms, Proceedings of SPIE Volume 8657 (Computational Imaging XI), Burlingame, CA, 2013
-
Jarrell Waggoner, Jeff Simmons, Marc De Graef, and Song Wang. Graph Cut Approaches for Materials Segmentation Preserving Shape, Appearance, and Topology. International Conference on 3D Materials Science (3DMS), 147-152, Seven Springs, PA, 2012.
-
Jarrell Waggoner, Jeff Simmons, and Song Wang. Combining global labeling and local relabeling for metallic image segmentation. In Proceedings of SPIE (Computational Imaging X), volume 8296, Burlingame, CA, 2012.
-
Song Wang, Jarrell Waggoner, and Jeff Simmons. Graph-cut methods for grain boundary segmentation. JOM Journal of the Minerals, Metals and Materials Society, 63(7):49–51, 2011.
Jarrell Waggoner
/-/ malloc47.com