Comparison of SOM Classification and Unsupervised KMeans image classification in SEXTANTE

The aim of this proof of concept exercise is to compare the results of two unsupervised classification modules from OTB: SOM Classification and Unsupervised KMeans image classification. The same input image used in our previous post has been used.

1. Perform the SOM Classification. The parameters are set as: Training set size to 20, Streaming lines to 1 and Number of iterations to 25.

2. The next step was to overlay this result with the result from the Unsupervised KMeans image classification with GRASS v.overlay and Operator to use set to not.

It is clear that SOM Classification classifies a larger number of objects.

a) Results of KMeans image classification


b) Results of SOM Classification


c) Difference