Metric and transform performance analysis for hyperspectral superpixel segmentation

Hyperspectral imaging is based on the acquisition of a large number of narrowly spaced spectral band images in the electromagnetic spectrum. Hyperspectral images surpass other imaging techniques in the detection of objects, classification and the detection of the changes that occur in the scene. A recent approach for hyperspectral segmentation is the superpixel segmentation approach. In this work, simple linear iterative clustering based superpixel segmentation method is adapted to hyperspectral images, and the effect of various metrics such as Euclidian distance, spectral information divergence, spectral angular distance, and the principal component analysis and minimum noise fraction transforms on the superpixel segmentation performance are evaluated.