Markov random field based image segmentation with adaptive neighborhoods to the detection of fine structures in SAR data

In the Markov random field (MRF) region label approach for synthetic aperture radar (SAR) image segmentation small structures may be lost. This is due to the filtering effect of the MRF region label model, which is desirable in homogeneously labelled areas like agricultural regions. End-users also interested in resource management, may wish to preserve the small structures such as small roads and rivers. To this end, the neighborhood set used in the MRF region label model has been made adaptive, based on a simple Bayesian network (BN). Results using synthetic aperture radar (SAR) data show that an important improvement of the representation of small structures is possible if they can be detected to some extent using the maximum likelihood approach.