Visualization of feature spaces based on spectral and texture characteristics

Abstract

The paper presents a method for visualizing feature spaces describing the pixels of hyperspectral images based on spectral and texture characteristics. The proposed method allows using various measures of spectral dissimilarity between image pixels along with textural features. The method is based on an implicit transition to intermediate feature representations using the calculation of pairwise dissimilarities between data points in input feature spaces, followed by the reconstruction of feature vectors in 3D space and interactive data visualization. The proposed approach is investigated on publicly available hyperspectral scenes using quantitative estimates and visual analysis.