Image Processing in Python for 3D image stacks.
Image Processing in Python for 3D image stacks, or IMPPY3D, is a software repository comprising mostly Python scripts that simplify post-processing and 3D shape characterization of grayscale image stacks, otherwise known as volume-based images, 3D images, or voxel models. IMPPY3D was originally created for post-processing image stacks generated from X-ray computed tomography measurements. However, IMPPY3D also contains a functions to aid in post-processing general 2D/3D images.
Python was chosen for this library because of it is a productive, easy-to-use language. However, for computationally intense calculations, compiled codes and libraries are used for improved performance, such as well known libraries like Numpy and SciKit-Image. Compiled libraries internal to IMPPY3D were created using Cython. IMPPY3D was developed in an Anaconda environment with Windows 10 and Linux in mind, and suitable Anaconda environment files for these operating systems are provided to simplify the process of installing the necessary dependencies although some dependancy resolution may still be required.
Some of the highlighted capabilies of IMPPY3D include: interactive graphical user-interfaces (GUIs) available for many image processing functions, various 2D/3D image filters (e.g., blurring, sharpening, denoising, erosion/dilation), the ability to segment and label continuous 3D objects, precisely rotating an image stack in 3D and re-slicing along the new Z-axis, multiple algorithms available for fitting rotated bounding boxes to continuous voxel objects, image stacks can be converted into 3D voxel models suitable for viewing in ParaView, and voxel models can be represented as smooth surface-based models like STL meshes. Additional information and example scripts can be found in the included ReadMe files.
Synthetic X-ray computed tomography (X-ray CT) images can be created. To do so run the "./resources/generate_sample_data.py" script, after follwoing the setup instructions. Some examples may have been downloaded as part of the code, these are found in the "./examples/resources/powder_particles/" directory. This image stack imitates what actual X-ray CT data would look like of isolated, metal powder-particles commonly used in metal-based additive manufacturing.
Using IMPPY3D, the synthetic image stack can be denoised and segmented. Then, the segmented particles can be converted into voxel model. As a demonstration, these powder particles were characterized in terms of volume, porosity, orientation, aspect ratio, sphericity, and more. Additional information can be found in the example script, "./examples/segment_3d_particles/segment_3D_powder_particles.py".
The development of IMPPY3D uses the Mamba
package manager for handling dependencies, similar to the popular Conda package
manager. We recommend creating a new Python 3.10 environment (in Mamba/Conda)
prior to installing IMPPY3D. The easiest method of installing IMPPY3D is
through pip. Wheel files(.whl) for pip can be found either on PyPi or in
the IMPPY3D repository(for Windows or Linux machines). Alternatively, IMPPY3D
can be compiled using setuptools. The following subsections go into more
details for each of these installation cases.
The simplest method of installing IMPPY3D is through PyPi.
Installing IMPPY3D from PyPi can be achieved using pip via the following,
pip install imppy3d==1.2.0
It is important that you explicitly specify the latest version of IMPPY3D, in this case, version 1.2.0. Moreover, the pip installation process of IMPPY3D is currently restricted to Python 3.10 environments.
The binary .whl files are located in the folder, "./dist/", of the
GitHub repository. The name
of the .whl files will contain information about the Python version, IMPPY3D
version number, and operating system. Currently, precompiled .whl files are
only available for Windows and Linux operating systems using Python 3.10. To
install IMPPY3D using one of these precompiled .whl files, choose the
appropriate .whl for your operating system and use the pip command in your
Python environment,
pip install /path/to/file.whl
IMPPY3D is largely Python-based, but there are also C extensions, via Cython,
that must be compiled in order to use all of the features of IMPPY3D. A
setup.py file is provided in the GitHub repository for easier compilation of IMPPY3D. To
compile IMPPY3D, open a terminal with your Python environment active and type,
python setup.py bdist_wheel sdist
This command will use the setuptools library to compile IMPPY3D into a wheel
(.whl). Upon successful completion, the binary .whl file will be located in the
folder, "./dist/". Additionally, the above command will create source
distribution, which is an alternative installation file to the binary
distribution. With the new .whl file created, simply follow the instructions
above stated about installing IMPPY3D using pip and a local binary file.
It is important to ensure that your Python environment also has the appropriate environment variables set for your C compiler. If setuptools cannot find your compiler, then compilation of IMPPY3D will fail. Furthermore, the same C-compiler as that which was used to compile the specific Python version you are using should also be used to create the IMPPY3D C extensions. For Windows users, see the documentation, https://wiki.python.org/moin/WindowsCompilers.
If you would like verify your installation of IMPPY3D, there are unit test scripts available for download in the "./tests/" folder of the GitHub repository. To execute all of the tests, run the script, "./tests/main_run_tests.py", which will print a text summary of the results to the console.
Additionally, in the GitHub repository, there are a number of example Python scripts provided in the "./examples/" folder. These example scripts serve as an excellent starting point when developing new post-processing scripts. As we continue to use IMPPY3D in new applications, we aim to continue to provide new example scripts in this folder.
It is well known that a stack of images can represent a 3D volume, but the coordinate systems in such data files is not standardized. IMPPY3D regularly switches between two coordinate systems: indicial coordinates and Cartesian coordinates. In IMPPY3D, an image stack is stored as 3D Numpy array, and so, the indices of this 3D array can directly serve as coordinates. However, it is often useful to use Cartesian coordinates when applying 3D rotations or visualizing the volume. When converting to Cartesian coordinates from indicial coordinates, care must be taken so that the resultant X-, Y-, and Z-axes in the Cartesian system define a right-handed coordinate system.
IMPPY3D assumes the following conversion between coordinate systems applies:
-
Ascending image indices corresponds to ascending Z-coordinates.
-
Ascending row indices corresponds to ascending Y-coordinates.
-
Ascending column indices corresponds to ascending X-coordinates.
For a given image, the 2D origin is taken to be the top-left corner of the image. Therefore, the X-axis points to the right (across the columns), and the Y-axis points downward (across the rows). However, the Z-axis in an image stack must be carefully considered. In IMPPY3D, we assume that the first image in the image stack corresponds to the TOP of the physical volume; every subsequent image represents a slice of the volume beneath the first image. In doing so, a right-handed coordinate system is naturally created based on the conversion of the indicial coordinates, as shown in the picture below.
Tomographic reconstructions from X-ray radiographs often result in the first
image in the image stack representing the bottom of the physical object. In
this case, we recommend reversing the image stack prior to any post-processing
in IMPPY3D. If the image stack is not reversed, then the resultant 3D visuals
(as .vtk files) will be reflections of the physical object. Reversing the image
stack can be automatically performed for you by using the 'flipz' parameter
when importing an image stack using imppy3d.import_export.load_image_seq().
-
Convert the comment blocks in function definitions to a common standard for automatic generation of the documentation using Sphinx.
-
Incoporate additional libraries like TomoPy for reconstruction of X-ray CT radiographs and removal of ring artifacts.
-
Create an optimization routine that stitches multiple X-ray CT fields-of-view together.
If you'd like to contribute to IMPPY3D, start by searching through the pull requests to see whether someone else has raised a similar idea or question. If you don't see your idea listed, and you think it fits into the goals of this project, open a pull request.
If you encounter any bugs or unintended behavior, please create an "Issue" in the IMPPY3D GitHub repository and label it with the "bug" tag. Provide information about the bug, and if possible, how to recreate the bug.
You can also make a request for new features using the Issues mechanism. Please label these requests with the "enhancement" tag.
For questions on how best to use IMPPY3D for a specific application, feel free to contact Dr. Newell Moser (see contact email below).
- Dr. Newell Moser, NIST ([email protected])
-
Dr. Alexander K. Landauer, NIST
-
Dr. Orion L. Kafka, NIST
-
Dr. Edward J. Garboczi, NIST Fellow (Retired)
-
The JOSS team and reviewers: @gknapp1, @sitic, and @danielskatz.
If IMPPY3D has been significant in your research, and you would like to acknowledge the project in your academic publication, we suggest citing the following NIST data repository:
Moser, Newell H., Landauer, Alexander K., Kafka, Orion L. (2023), IMPPY3D: Image processing in python for 3D image stacks, National Institute of Standards and Technology, https://doi.org/10.18434/mds2-2806