Fusion of spectral and texture features for hyperspectral data clustering

In this paper, we apply the recently proposed framework for the fusion of spectral and texture features in a hyperspectral data clustering problem. The proposed pipeline allows us to use different spectral dissimilarity measures. It starts with the calculation of features, followed by the implicit transition to an intermediate feature representation with the calculation of the pairwise distances between data points in space of fused features. The next stages include the reconstruction of feature vectors in low-dimensional space followed by a data clustering. We study the proposed approach using well-known hyperspectral scenes and compare the results to classical hyperspectral image segmentation approaches.