Fusion of LBP and hu-moments with fisher vectors in remote sensing imagery

Abstract

There are huge volumes of scene images generated periodically by satellite technology which require effective processing for intelligent decisions. The satellite sensed images contain diverse image contents such as shape, color, texture, spectral, and spatial resolutions. These variables are further affected with varying illumination conditions which result to noisy images. This poses a challenge to image analysis methods thus limiting their capabilities tasks of image understanding and interpretation. Remote sensing imagery classification task hence need advanced methods which can characterize images better so as to achieve higher accuracy results on scene images classification. This research proposes a feature-fusion-strategy of complementary image features, this is, Local Binary Patterns (LBP) and Hu Moments are fused with fisher vectors and the resultant is more discriminative feature representation that yield better image scene classification accuracy of 50.12% as demonstrated in with experimental results. This is an improvement compared to some pixel-based image descriptor methods in literature that are implemented. Although there are other methods in literature with superior classification accuracies than the proposed method, its is evident that fusion of complementary-features result to better classification accuracy.