With the same set of wood image data, a SOM (self-organizing map, Kohonen) was trained. An introductory description and the code for this example was originally taken from here.

WoodSOM

SOM of 54 wood samples

Every sample consists of a 16 x 16 (256) pixels image, every pixel having three dimensions for red, green and blue respectively. Every Node of the map hosts therefore a 768-dimensional vector.

The number 920 in the top left corner indicates the number of training iterations passed of a total of 2000. After all the runs have finished, the clusters around every sample show much clearer borders.