A Metrological Hyperspectral Texture Similarity Measure using Kullback-Leibler Divergence

Abstract

t A metrological hyperspectral texture descriptor, Relative Spectral Difference Occurrence Matrix (RSDOM) has been proposed in previous work. Due to the probabilistic nature of the feature, Kullback-Leibler divergence (KLD) is used for similarity measure. In theliterature, several approaches exist to approximate KLD which is otherwise computationallyexpensive if processed directly for feature spaces of high dimensionality. In this work, wecompare the performance of these approaches through a hyperspectral texture classificationscheme. The result shows close performance of variational approach (using Gaussian mixturemodel) to direct processing (using binned histogram) with 600 % faster speed. Key wordsTexture, non-uniformity, hyperspectral, metrology, Kullback-Leibler divergence.