Hyperspectral image super-resolution algorithm via sparse representation based on spectral similarity

Hyperspectral image sparse super-resolution algorithm based on spectral similarity was proposed to improve low spatial resolution of hyperspectral images. The super resolution algorithm, based on the criterion of maximum likelihood estimation and Gaussian mixture sparse representation, assigned various weights to different coding residuals to improve spatial resolution of reconstructed images and the robustness to noise. Based on spectral similarity, the super-resolution model which added sparsity constraints using pixel spectral similarity was proposed to ensure the accuracy of the spectrum images. The experiments have been run to prove that this model achieves a better result than Bicubic, Yang and Pan algorithms in both visual effect and objective measures. Additionally, various parameters in the reconstruction were analyzed in order to provide better image detection and classification.