Dimension selective tensor compression of hyperspectral images

An efficient method for hyperspectral image compression is presented using tensor approximation. Hyperspectral images are first modeled as 3D tensors. Every tensor is then represented using its Tucker representation and matrices for every mode are calculated. Choosing eigenvectors corresponding to greatest eigenvalues of projection matrices, we reach a lower order tensor. Our method not only exploits redundancies between bands but also uses spatial correlations of every band image and therefore, as simulation results applied to airborne visible/infrared imaging spectrometer (AVIRIS) files demonstrate, leads to a remarkable compression ratio and quality.