segmentation in the feature space

There are at least two nice ways to accomplish this. First, you can use oft-seg, pass the output of the oft-clump to the program using the -um option and give a desired minimum size in pixels. Second, you can use the gdal_sieve.py. The difference between the two is that whereas oft-seg merges the small segments with the most similar neighboring segment gdal_sieve.py merges them with the largest neighboring polygon.