Spectral-Spatial Compressive Sensing of Hyperspectral Images with Spectral Clustering and Reconstruction via Spectral-Spatial Total Variation Regularization

Authors

School of Electrical Engineering, IUST, Tehran, Iran

Abstract

In this paper, considering the correlation of spectral bands of a hyperspectral image, first we cluster these bands based on correlation coefficients. Then, using spatial correlation among the pixels of a hyperspectral image and the mentioned clustering, we propose a spectral-spatial compressed sensing for hyperspectral images. For reconstruction of these images, we propose a spectral-spatial total variation regularization in which in addition to the vertical and horizontal discrete gradients, we incorporate the frequency discrete gradient as well. Using the mentioned clustering, reconstruction computations of spectral bands of clusters can be performed in parallel leading to a higher reconstruction speed. Also, in the case of spectral-spatial compressed sensing without clustering and performing reconstruction, the proposed method in comparison to the norm based and spatial total variation regularization methods improves the reconstruct quality in terms of PSNR. Computer simulations on actual hyperspectral images confirm the above results.

Keywords

TABRIZ JOURNAL OF ELECTRICAL ENGINEERING
Volume 47, Issue 4 - Serial Number 82
February 2017
Pages 1513-1521

Files

Share

How to cite

Statistics