Spatial-spectral Dictionary Learning And Hyperspectral Image Reconstruction

Hyperspectral images have significant structural features, If the hyperspectral imagecan be properly characterized, data collection and data analysis can be improved. Becausemost of the pixels, as reflected in the spectral reflectance curves of several materials just a few,so we think the sparse coding model and hyperspectral image data is a good match. Sparsemodel considers each pixel is just a combination of several large dictionary of elements, andin the application of this method has been proven very effective.A spatial-spectral dictionary learning algorithm is introduced and applied to reconstructthe hyper-spectral images. According to the characteristic that hyper-spectral images haverich spatial and spectral correlations, the hyper-spectral images can be divided into3D smallcube blocks. Therefore, we use this dictionary to describe these cube blocks. we use agradient descent method to learning dictionary. First, assuming the dictionary is fixed,gradient descent method is used to calculate the sparse coefficients; Second, assuming thecoefficients are fixed, gradient descent method to update the dictionary; these two steps arealternately used until the algorithm converges. Finally, the dictionary is applied to reconstructhyper-spectral images, Experimental results show that the method can obtain a goodreconstruction results by comparing the values of PSNR.