Local Descriptors Parameter characterization with Fisher vectors for remote sensing images

Satellite technology yield huge quantities of high spatial resolution(HSR) images periodically. The HSR type of data is complex in its spatial arrangement with high intraclass and low interclass variability. Remote sensing images scene classification is a challenging task due high inter and intra class variations due to diverse scene contents, induced noise as a resultant of changes in illuminations, differing scales and rotations of images. Consequently, no one-specific image descriptor algorithm is effective to characterize scene image-semantics for accurate classification. This research employ Fisher vector to characterize parameters of local descriptors i.e. Local Ternary Patterns (LBPs) and Hu Moments to a high fisher-vector-feature-representation that is more discriminative for remote sensing image scene classification. Support Vector Machine Classifier is implemented to validate the result. Overall results of 52.29 is achieved using the proposed strategy show a significant improvement compared to individual image descriptor algorithms in literature.