Premium
Compressed sampling reconstruction of hyperspectral images based on spectral prediction
Author(s) -
Wang Li,
Feng Yan
Publication year - 2016
Publication title -
ieej transactions on electrical and electronic engineering
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.254
H-Index - 30
eISSN - 1931-4981
pISSN - 1931-4973
DOI - 10.1002/tee.22189
Subject(s) - hyperspectral imaging , sampling (signal processing) , artificial intelligence , pattern recognition (psychology) , full spectral imaging , correlation coefficient , spectral bands , computer science , residual , compressed sensing , iterative reconstruction , mathematics , algorithm , computer vision , remote sensing , statistics , filter (signal processing) , geology
With increasing amounts of hyperspectral images (HSI) and the limitations of the memory requirements, compressive techniques for hyperspectral images have attracted extensive research efforts in recent years. The main difficulty of applying compressed sampling (CS) theory to compression and reconstruction of hyperspectral images lies in using the spatial correlation and spectral correlation of hyperspectral images. In this paper, a reconstruction algorithm of hyperspectral images taking advantage of two‐dimensional compressed sampling (2DCS) and two‐dimensional total variation (2DTV) incorporating spectral prediction (SP) is investigated. In the sampling process, the hyperspectral images are divided into reference bands and common bands, and all bands are sampled using 2DCS independently. In the reconstruction process, the reference bands are reconstructed by 2DTV first. In order to improve the reconstruction quality of common bands, spectral prediction utilizing the spectral correlation between reference bands and common bands is conducted. Then the spectral compensation is computed by using a combination of the prediction value and the initial approximation for the common bands. The residual between the compensation value and the original value is obtained to revise the approximation for the common bands. The algorithm is implemented in an iterative manner to enhance the performance. Experimental results on popular hyperspectral datasets reveal that the proposed algorithm exploiting spectral prediction outperforms the algorithm 2DCS‐2DTV, which does not use spectral correlation, as well as the state‐of‐the‐art algorithms in terms of peak signal‐to‐noise ratio (PSNR). In particular, when the sampling rate of the reference bands is higher than that of the common bands, the proposed algorithm would improve the reconstruction quality dramatically. © 2015 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.