An iterative approach to partially supervised classification problems

A novel partially supervised classification technique is proposed, which allows the efficient mapping of a specific land-cover class (or a few land-cover classes) of interest, by using only training samples belonging to the class or classes selected. It is based on a combined use of a Radial Basis Function network, which models the image data distribution, and a Markov Random Field approach, which exploits the spatial-contextual information. The result is high classification accuracy comparable to that provided by fully supervised classifiers.