Optimized class-separability in hyperspectral images

Image visualization techniques are mostly based on three bands as RGB color composite channels for human eye to characterize the scene. This, however, is not effective in case of hyper-spectral images (HSI) because they contain dozens of informative spectral bands. To eliminate redundancy of spectral information among these bands, dimensionality reduction (DR) is applied while at the same trying to retain maximum information. In this paper, we propose a new method of information-preserved hyperspectral satellite image visualization that is based on fusion of unsupervised band selection techniques and color matching function (CMF) stretching. The results show consistent, edge-preserved and pre-attentive feature less images with high class separability. Different visualization techniques are compared to demonstrate the effectiveness of our scheme that can prompt an important advancement in the field.