LBP and Weber law descriptor feature based CRF model for detection of man-made structures

In this paper, we have proposed a combined Local Binary Pattern (LBP) and Weber Law Descriptor (WLD) feature based Conditional Random Field (CRF) model for detection of man made structures such as buildings in natural scenes. In natural scenes, the structure may have textural attributes or some portions of the object may be apparent as textures. The CRF model learning has been carried out in feature space. The spatial contextual dependencies of the structures has been taken care by the intrascale LBP features and interscale WLD features. The CRF model learning problem have been formulated in pseudolikelihood framework while the inferred labels have been obtained by maximizing the posterior distribution of the feature space. Iterated conditional mode algorithm (ICM) has been used to obtain the labels. The proposed algorithm could successfully be tested with many images and was found to be better than that of Kumar´s algorithm in terms of detection accuracy.