2.36. Image clustering

Performs automatic clustering of an image (e.g. satellite image) into a specified number of classes. K-Means algorithm is used

Parameters that can be set up:

  • Number of clusters into which the original image will be divided.

  • Method of K-Means initialization. Could be ‘k-means++’ or ‘random’. Leave empty to use default value (k-means++).

‘k-means++’ : selects initial cluster centroids so as to make them furhter apart. This technique speeds up convergence.

  • Maximum number of iterations of the k-means algorithm for a single run. Leave empty to use default value (300).

The necessary number of iterations depends on many factors including centroid characteristics and number of clusters. Usually between several dozens and several hundreds iterations are enough.

  • K-means algorithm to use. Could be one of ‘lloyd’, ‘elkan’. Leave empty to use default value (lloyd).

Lloyd’s algorithm is the classical EM-style algorithm. The “elkan” variation can be more efficient on some datasets with well-defined clusters, by using the triangle inequality. However it’s more memory intensive.

Input:

  • Image in GDAL-compatible format (preferably GeoTIFF). It can contain any number of bands, all of which will be used for cluster calculation.

Output:

  • Clustered image.

Launch tool: https://toolbox.nextgis.com/operation/image_clustering

Try it out using our sample:

Download input dataset to test the instrument. Step-by-step instructions included.

Get the output to additionally check the results.