Simple (or more fittingly: stupid) image clustering [sic] is a very barebones command line tool written in Python 3.
sic requires opencv2 (and also numpy) to perform clustering and manipulate files. Matplotlib is needed for additional plotting.
To use this command line tool simply clone it into any directory you want:
$ git clone https://github.com/butkej/simple-image-clustering.git
Afterwards, simply change into the downloaded directory, make the script executable and run it:
$ cd simple-image-clustering/
$ chmod +x install.sh
$ ./install.sh
This will pip install the package into the currently activated environment and makes the script executable by calling only sic itself with its respective arguments.
$ (python) sic(.py) [-h] -i INFILE [-s SPECTRAL_SWITCH] [-c NUM_CLUSTERS] [-r NUM_RUNS] [-e EXTENSION] [--version]
simple image clustering based on k-means clustering
optional arguments:
-h, --help show this help message and exit
-i INFILE, --infile INFILE you must provide an input image to cluster it
-s SPECTRAL_SWITCH, --spectral SPECTRAL_SWITCH switch to spectral clustering mode (see documentation for details) (FOUND BELOW IN THE README)
-c NUM_CLUSTERS, --clusters NUM_CLUSTERS choose the number of clusters to produce (default is 5) [integer]
-r NUM_RUNS, --runs NUM_RUNS choose the number of clustering runs to attempt (default is 10) [integer]
-e EXTENSION, --extension EXTENSION choose the filename extension of the saved clustered image (default is png) [string]
The spectral option takes an input folder (containing single wavenumber images) instead of a single input image!
All .tif images in this folder will be loaded sequentially, converted into grayscale images and concatenated.
This new array is then clustered to achieve a clustering across all spectral wavelenghts in the folder.
The clustered image is saved by converting cluster labels to a distinct color palette. This leads to a maximum of 21 clusters (Red is reserved for merging as cluster color 22).
By examining the saved clustering image the user can then choose a suitable cluster (prompted during script execution). The spectra from this cluster are then extracted, plotted and saved as well.
Additionally all cluster spectra are saved to an extra folder individually.
After computation of the first false color image and the saving of all mean spectra of all clusters the user is asked if he wants to merge clusters.
Here, each cluster that should be merged can be given or just press enter to abort further clusters. These clusters are then merged and set to a new color (=RED) in the new false color image. Spectra are extracted and saved again.