Bilateral texture filtering for spectral-spatial hyperspectral image classification

• Main Authors: Ying Zhang, Jing He
• Format: Article
• Language: English
• Published: Wiley 2019-12-01
• Series: The Journal of Engineering
• Subjects: image texture; feature extraction; image classification; hyperspectral imaging; support vector machines; geophysical image processing; bilateral texture; spectral-spatial hyperspectral image classification; novel structure-preserving filtering based method; hyperspectral images; authors partition; subsets; neighbouring bands; resulting features; btf; fused bands; structure-preserving filtering technology; removing texture; main structure information; source image; filtered image; classification result; popular hyperspectral data sets
• Online Access: https://digital-library.theiet.org/content/journals/10.1049/joe.2018.9211

Here, a novel structure-preserving filtering based method is proposed for feature extraction of hyperspectral images (HIs). In the first step, the authors partition the HI into several subsets of neighbouring bands and then fuse the bands in each subset by averaging method. Second, the resulting features are obtained by bilateral texture filtering (BTF) on the fused bands. BTF is a structure-preserving filtering technology, which aims at removing texture while preserving main structure information of source image. Finally, the SVM classifier is performed on the filtered image to obtain the classification result. Experiments tested on two popular hyperspectral data sets show that the proposed method outperforms some other widely used methods.