Spectral Representation vis Data-Guided Sparsity for Hyperspectral Image Super-Resolution†.

Xian-Hua Han,Yongqing Sun,Boxin Shi,Yen-Wei Chen,Jian Wang,Yinqiang Zheng

doi:10.3390/s19245401

Abstract

Hyperspectral imaging is capable of acquiring the rich spectral information of scenes and has great potential for understanding the characteristics of different materials in many applications ranging from remote sensing to medical imaging. However, due to hardware limitations, the existed hyper-/multi-spectral imaging devices usually cannot obtain high spatial resolution. This study aims to generate a high resolution hyperspectral image according to the available low resolution hyperspectral and high resolution RGB images. We propose a novel hyperspectral image superresolution method via non-negative sparse representation of reflectance spectra with a data guided sparsity constraint. The proposed method firstly learns the hyperspectral dictionary from the low resolution hyperspectral image and then transforms it into the RGB one with the camera response function, which is decided by the physical property of the RGB imaging camera. Given the RGB vector and the RGB dictionary, the sparse representation of each pixel in the high resolution image is calculated with the guidance of a sparsity map, which measures pixel material purity. The sparsity map is generated by analyzing the local content similarity of a focused pixel in the available high resolution RGB image and quantifying the spectral mixing degree motivated by the fact that the pixel spectrum of a pure material should have sparse representation of the spectral dictionary. Since the proposed method adaptively adjusts the sparsity in the spectral representation based on the local content of the available high resolution RGB image, it can produce more robust spectral representation for recovering the target high resolution hyperspectral image. Comprehensive experiments on two public hyperspectral datasets and three real remote sensing images validate that the proposed method achieves promising performances compared to the existing state-of-the-art methods.

Highlights

Hyperspectral (HS) imaging is an emerging technique for simultaneously obtaining a set of images of the same scene on a large number of narrow band wavelengths
Motivated by the fact that the material purity of each pixel spectrum might be different from the others, we firstly explore the possible purity of each pixel according to the local content similarity of a focused pixel and incorporate the material purity as the sparsity constraint into the non-negative sparse spectral representation, which adaptively imposes the sparse constraint for each pixel
G-SOMP+ could achieve better performance than the fixed number of the used dictionary atoms for spectral representation, and our proposed method with data guided sparsity could further improve the performance for most hyper-parameters and all quantitative metrics in the CAVE dataset

Summary

Introduction

Hyperspectral (HS) imaging is an emerging technique for simultaneously obtaining a set of images of the same scene on a large number of narrow band wavelengths. HS imaging can provide high spectral resolution, it imposes a severe limitation on the spatial resolution compared with the general RGB cameras. In order to guarantee a sufficient signal-to-noise ratio, enough exposure amount is needed for each narrow wavelength window, which can be generally solved in the existing hyperspectral cameras via collecting exposures in a much larger spatial region than the common RGB cameras, resulting in much lower spatial resolution. The low spatial resolution may result in the possible spectral mixture of different materials and restricts its performances for scene analysis and understanding. The high spatial resolution multi-spectral (e.g., RGB and RGB-NIR) images are available for the same scene with the common color cameras. The fusion method can effectively utilize the spectral correlation property in the LR-HS image and the detailed spatial structure in the HR-RGB image and generate a more accurate

Objectives

Results

Conclusion