This project has mainly migrated to https://github.com/Eve-ning/glcm-cupy which has many more new features & optimization improvements.

This repository will be left here for archival purposes.

Photo by Flo Maderebner from Pexels

400 Times faster, the above image takes 6+ Days to process, compared to 24 Minutes with glcmbin5 with bins=16. YMMV

I recommend to install this via forking. The pip version may be unstable.

Run this command with the c_setup.py here

python c_setup.py build_ext --inplace

• Implemented in Cython, means you get C-Lang Speed on Python
• 2/4/6/8-directional GLCM, in the case where orientation doesn't matter
• Memory Optimized, calculations are per window, thus a large intermediate GLCM isn't used, saving GBs of memory.

Pitfalls

This doesn't work with parallel processing for some reason, I couldn't get it to work unfortunately.

If you derived this work to make it parallel, please let me know! I would love to use it.

This is ~400 times faster than using skimage.feature.graycomatrix and graycoprops because this is Cython optimized, YMMV.

With a 2000x1000x3 image, it takes around 6+ minutes. Compared to 39 hours with skimage

GLCM Progress: 100%|██████████| 12/12 [06:49<00:00, 34.09s/it]

Binning the input array reduces size of GLCM, also increases performance. Though there's no substantial evidence of it improving nor deproving results.

With this algorithm, I omit generating the whole GLCM, instead, it's integrated in the GLCM feature calculation. Memory used is freed asap.

You can also see an example in /examples

from glcm.glcm import CyGLCM
import numpy as np
ar = ...
glcm = CyGLCM(ar.astype(np.float32),
              radius=3,
              bins=8,
              pairs=('H', 'V', 'SE', 'NE')
              ).create_glcm()

• radius: The radius of the GLCM window
• bins: The number of bins to use
• pairs:
  • H: Horizontal Pair
  • V: Vertical Pair
  • SE: South-East Diagonal Pair
  • NE: North-East Diagonal Pair

Input:

• ndim = 3
• shape=(in_dim0, in_dim1, channel)

Output:

• ndim = 4
• shape=(in_dim0, in_dim1, channel, features)
• Methods:
  • Homogeneity
  • Correlation
  • Angular Second Moment
  • GLCM Mean
  • GLCM Variance

The progress bar value is the current pair calculated.

The resulting GLCM array will be smaller than the original.

GLCM Dimension = Dimension - (2 * radius + step_size) = Dimension - Diameter

Arrays MUST BE in np.float32, you need to cast it.

ar.astype(np.float32)

Based on GLCM Texture: A Tutorial v. 3.0 March 2017.

For an effective segmentation, we just need 5 features as selected here.

Many features are not significantly orthogonal, hence more will introduce redundancy.

Chosen methods are for simplicity and efficiency in coding.

0.1.5 swapped Contrast for Homogeneity as it may be more orthogonal.

Arrays are Binned before going through GLCM.

All arrays will be processed to integer values [0,bin-1] band-independently.

If you have used or referenced any of the code in the repository, please kindly cite

@misc{glcmbin5,
  author = {John Chang},
  title = {Binned Cython 5 Feature GLCM},
  year = {2021},
  publisher = {GitHub},
  journal = {GitHub Repository},
  howpublished = {\url{https://github.com/Eve-ning/glcmbin5}},
}

• Dr. Ji-Jon Sit for hosting the parent project Eve-ning/FRModel which lead to this algorithm
• Wang Ji Fei for discovering GLCM Binning optimization.
• GLCM Texture: A Tutorial v. 3.0 March 2017 for providing a simple tutorial to guide this implementation.