Hyperspectral Image Segmentation by Self Organized Learning-Based Active Contour Model

Abstract

The growing attention to the hyperspectral sensors is driven by their special ability to provide rich information about various objects in a scene like surface minerals, water, snow, vegetation, pollution, man-made objects, etc.; enabling effective object segmentation. In this paper, we present a hyperspectral image segmentation methodology that incorporates the local hyperspectral information into a learning-based active contour level-set function for an accurate object region and boundary extraction. The segmentation process is achieved by utilizing self-organized lattice Boltzmann active contour (SOLBAC) technique that is based on constructing a self-organized Local Image Fitting (LIF) level-set cost function, for accurate and fast boundary extraction. The proposed algorithm starts with feature extraction from raw hyperspectral images that leverages the principal component analysis (PCA) transformation to reduce dimensionality and select the best sets of the significant spectral bands. Then, the SOLBAC approach is applied on the optimal number of spectral bands determined by the PCA. By using the properties of the collective computational ability and energy convergence capability of the Lattice Boltzmann Method (LBM), our proposed segmentation is capable of producing faster segmentation by more than 30% when compared to the state-of-the-art segmentation methods. The LBM is adopted for faster curve evolution of the level-set function and to stop the evolution of the curve at the most optimum object region. Experiments performed on our test dataset show promising results in terms of time and quality of the segmentation when compared to other state-of-the-art learning-based active contour model approaches.