Universal high spatial resolution hyperspectral imaging using hybrid-resolution image fusion

This is a color space transfer-based super-resolution algorithm for HSI-RGB fusion in coping with unknown spectral degradation. By transferring the HR-RGB to CIE XYZ color space, we employ the CIE color matching function (CMF) as the spectral degradation to reconstruct the HR-HSI, which successfully skip the SRF measurement. To efficiently fuse the HR-XYZ and the LR-HSI, we propose a polynomial fusion model that estimates the ratio matrix between the target HR-HSI and the upsampled LR-HSI. The target HR-HSI is reconstructed by combining the ratio matrix and the unsampled LR-HSI. The quantitative results outperform exisiting SOTA (2014-2021) algorithms.

Flowchart

Abstract

By fusing a low spatial resolution hyperspectral image (LR-HSI) and a high spatial resolution RGB image (HR-RGB), hybrid-resolution hyperspectral imaging has been a popular framework for acquiring high spatial resolution hyperspectral image (HR-HSI). Existing fusion methods always employ a known spectral response function (SRF) of the RGB camera to reconstruct the HR-HSI. The SRF is often limited or unavailable in practice, restricting the performance of existing methods. To address this problem, we propose a color space transfer-based fusion strategy that obtains HR-HSI based on a hybrid resolution hyperspectral imaging system without measuring SRF. Specifically, by using clustered-based back propagation neural network (BPNN), the HR-RGB is mapped into the CIE XYZ color space, and the HR-XYZ is obtained. In the CIE XYZ color space, its SRF is known, which successfully skip the SRF measurement. To efficiently fuse the HR-XYZ and the LR-HSI, we propose a polynomial fusion model that estimates the ratio matrix between the target HR-HSI and the upsampled LR-HSI. Finally, the target HR-HSI is reconstructed by combining the ratio matrix and the unsampled LR-HSI. Experimental results on two public data sets and our real-world data sets show that the proposed method outperforms five state-of-the-art fusion methods.

Result presentation

Guidance

Add your dataset path in config.py
Run main.py

Requirements

Environment

Python3.8
torch 1.12,torchvision 0.13.0
Numpy,Scipy,opencv-python,scikit-learn

Datasets

CAVE dataset, Preprocessed CAVE dataset.

Note

For any questions, feel free to email me at [email protected].
If you find our work useful in your research, please cite our paper ^.^

Citation

@article{article,
author = {Cao, Xuheng and Lian, Yusheng and Liu, Zilong and Zhou, Han and Bin, Wang and Huang, Beiqing and Zhang, Wan},
year = {2023},
month = {03},
pages = {003107},
title = {Universal high spatial resolution hyperspectral imaging using hybrid-resolution image fusion},
journal = {Optical Engineering},
doi = {10.1117/1.OE.62.3.033107}
}