hvn2 / DIP-SLR

Codes for the paper "Deep sparse and low-rank for HSI denoising" in Proceeding of IGARSS 2022, Kuala Lumpur.

DIP-SLR

Offical Python codes for the paper "Deep sparse and low-rank for hyperspectral denoising", in Proceeding of IGARSS 2022, Kuala Lumpur, July 2022.
Authors: Han V. Nguyen $^\ast \dagger$, Magnus O. Ulfarsson , Jakob Sigurdsson $^\ast$ and Johannes R. Sveinsson $^\ast$
$^\ast$ Faculty of Electrical and Computer Engineering, University of Iceland, Reykjavik, Iceland
$^\dagger$ Department of Electrical and Electronic Engineering, Nha Trang University, Khanh Hoa, Vietnam
Email: [e-mail]

Please cite our paper if you are interested
@inproceedings{nguyen2022deep, title={Deep Sparse and Low-Rank Prior for Hyperspectral Image Denoising}, author={Nguyen, Han V and Ulfarsson, Magnus O and Sigurdsson, Jakob and Sveinsson, Johannes R}, booktitle={IGARSS 2022-2022 IEEE International Geoscience and Remote Sensing Symposium}, pages={1217--1220}, year={2022}, organization={IEEE} }

Abstract:

Spectral and spatial correlation in hyperspectral images (HSIs) can be exploited in HSI processing because it directly induces a sparse and low-rank prior via linear transformations. Researchers have used sparse and low-rank prior as an image prior for HSI restoration, such as denoising, deblurring, and super-resolution. This paper proposes an HSI denoising method that incorporates a sparse and low-rank prior with a deep image prior (DIP). The sparse and low-rank prior is obtained using 2-dimensional discrete wavelet transform (2-D DWT), and singular value decomposition (SVD), while the DIP is provided by the structure of a convolutional neural network (CNN). The combination of a sparse and low-rank prior with a DIP views the CNN-based denoising method similar to a model-based method, inheriting the advantages of both model-based and CNN-based methods. Experimental results with simulated and real HSI datasets show that the proposed method outperforms the conventional sparse and low-rank based methods in both quantitative and qualitative performance.

Usage:

The following folders contain:

• data: The simulated (PU and DC) and the real data.
• models: python scripts define the models
• utils: additional functions
  Run the jupyter notebooks and see results.

Environment

• Pytorch 1.8
• pytorch_wavelets 1.3
• Numpy, Scipy, Skimage.