Discriminative random fields with belief propagation inference: Applications in semantic-based classification of remote sensing images

This paper addresses the problem of remote sensing image classification based on the semantic context using Discriminative Random Field (DRF) model. The DRF model is used to capture the highly complicated spatial interactions and contextual information in remote sensing images. The DRF labels different image regions by using neighborhood spatial interactions of the labels as well as the observed data. Based on the DRF model, a graph-based inference algorithm - Belief Propagation (BP), is employed to obtain the optimal classification result. This inference algorithm is efficient in the sense that it produces highly accurate results in practice compared to other traditional inference algorithms.