Segmentation of remote-sensing images by supervised TS-MRF

In this work we specialize the recently proposed tree-structured MRF model to supervised segmentation of multispectral satellite images. This model allows a hierarchical representation of a 2-D field by means of a sequence of binary MRFs, each corresponding to a node in the tree. One can fit the intrinsic structure of the data to this tree-structured model, thereby defining a multi-parameter, flexible, MRF. Although a global MRF model is defined on the whole tree, optimization as well as estimation is carried out node by node, with a significant reduction in complexity. Experiments on a test SPOT image prove the superior performance of the algorithm w.r.t. other MRF-based or variational algorithms for supervised segmentation.