RGB Patch Clustering For Hyperspectral Image Super-resolution Using Sparse Coding

Hyperspectral images (HSI's) are characterized by low spatial resolution and high spectral resolution. In this paper, we propose a method for improving the spatial resolution of HSI's making use of the high spatial resolution RGB image. We solve this problem under a sparse reconstruction framework with clustering of similar patches in the high resolution RGB image. To this end, we first learn a dictionary of atoms from the given low resolution HSI, followed by clustering of similar patches in the RGB image. Sparse coefficients corresponding to individual clusters are estimated using Matching Pursuit Algorithm. Since RGB image is considered to be the spectrally transformed version of desired high resolution HSI, we use the coefficients estimated as above for the reconstruction of high resolution HSI. Patchwise clustering ensures the spatial similarity between various patches in a cluster. Experimental results show that the proposed method effectively recovers the spatial details, preserving the spectral information and removes the block artifacts associated with an independent application of GSOMP on all patches. Our method gives better result compared to other state of the art techniques like MF, SASFM and GSOMP.