Professional Documents
Culture Documents
Guided Hyperspectral Image Denoising With Realistic Data: Tao Zhang Ying Fu Jun Zhang
Guided Hyperspectral Image Denoising With Realistic Data: Tao Zhang Ying Fu Jun Zhang
https://doi.org/10.1007/s11263-022-01660-2
Abstract
The hyperspectral image (HSI) denoising has been widely utilized to improve HSI qualities. Recently, learning-based HSI
denoising methods have shown their effectiveness, but most of them are based on synthetic dataset and lack the generalization
capability on real testing HSI. Moreover, there is still no public paired real HSI denoising dataset to learn HSI denoising
network and quantitatively evaluate HSI methods. In this paper, we mainly focus on how to produce realistic dataset for
learning and evaluating HSI denoising network. On the one hand, we collect a paired real HSI denoising dataset, which
consists of short-exposure noisy HSIs and the corresponding long-exposure clean HSIs. On the other hand, we propose an
accurate HSI noise model which matches the distribution of real data well and can be employed to synthesize realistic dataset.
On the basis of the noise model, we present an approach to calibrate the noise parameters of the given hyperspectral camera.
Besides, on the basis of observation of high signal-to-noise ratio of mean image of all spectral bands, we propose a guided HSI
denoising network with guided dynamic nonlocal attention, which calculates dynamic nonlocal correlation on the guidance
information, i.e., mean image of spectral bands, and adaptively aggregates spatial nonlocal features for all spectral bands. The
extensive experimental results show that a network learned with only synthetic data generated by our noise model performs as
well as it is learned with paired real data, and our guided HSI denoising network outperforms state-of-the-art methods under
both quantitative metrics and visual quality.
Keywords Hyperspectral image denoising · Paired real dataset · Realistic synthetic data · Guided nonlocal denoising network
1 Introduction RGB image, and various noises are easily introduced into
the acquisition process. This kind of degradation negatively
Hyperspectral image (HSI) can provide much more spectral influences not only the visual appearance of the HSI but also
information than RGB image, and is beneficial to numerous the performance of all downstream HSI applications (Chen
applications, including remote sensing (Borengasser et al. et al. 2014). Thus, HSI denoising is an essential step in the
2007; Ojha et al. 2015), computer vision (Cao et al. 2018; pipeline of HSI analysis and processing (Fig. 1).
Kwon and Nasrabadi 2007), medical diagnosis (Lu and Fei To remove the imaging noise, the well-known model-
2014; Bjorgan and Randeberg 2015) and more. Hyperspec- based HSI denoising methods often iteratively solve an
tral imaging is to capture spectral information of each spatial optimization problem with various hand-crafted priors, such
position in a scene with massive wavebands, and the commer- as smoothness (Yuan et al. 2012), self-similarity (Fu et al.
cial hyperspectral cameras often utilize the scanning design 2017; He et al. 2019) and so on. Nevertheless, the iterative
(Porter and Enmark 1987; Basedow et al. 1995). These make optimization procedure is time-consuming and the hand-
its photon counts for each band are much less than that in crafted priors cannot sufficiently represent the variety of
data in the real world. Instead of costly optimization and
Communicated by Boxin Shi. hand-crafted priors, learning-based methods (Wei et al. 2021;
Chang et al. 2018; Yuan et al. 2018; Dong et al. 2019; Lin
B Ying Fu et al. 2019) automatically learn the mapping from noisy HSI
fuying@bit.edu.cn
to clean HSI with convolutional neural network (CNN). How-
1 School of Computer Science and Technology, Beijing ever, existing leaning-based methods rarely consider the high
Institute of Technology, Beijing 100081, China signal-to-noise ratio (SNR) data property of mean image of
2 Advanced Research Institute of Multidisciplinary Science, all spectral bands, as shown in Fig. 2, and dynamic spatial
Beijing Institute of Technology, Beijing 100081, China
123
International Journal of Computer Vision
123
International Journal of Computer Vision
123
International Journal of Computer Vision
123
International Journal of Computer Vision
123
International Journal of Computer Vision
where Y is the noisy HSI, N indicates the summation of all N p (n, m, λ) ∼ g(0, σ p2 (λ)), (4)
noises physically caused by light and camera, L represents
the number of photoelectrons that is proportional to the scene where g(·) represents the Gaussian distribution and σ p (λ)
irradiation, and k denotes the system gain, respectively. Note denotes the scale parameter for the λ-th band.
that we assume the system gain of all elements in the captured Apart from these signal-independent noise, due to spa-
HSI is identical due to the same CCD sensor in hyperspectral tial scanning design, hyperspectral camera often suffers from
camera. stripe pattern noise, as illustrated in Fig. 3. As we analyze
To systematically analyze the noise in HSI, we divide the hyperspectral camera employed to capture our real HSI
the noise into two components, e.g., signal-dependent noise dataset, we find it contains both horizontal stripe pattern noise
(correlated to incident light) and signal-independent noise and vertical stripe pattern noise, as shown in Fig. 7. The rea-
(uncorrelated to incident light). son is that each row is captured by the same CCD unit during
horizontal scanning, which causes the horizontal stripe pat-
4.1.1 Signal-Dependent Noise tern noise. Meanwhile, each column is captured in different
time, which causes the vertical stripe pattern noise. Thus, the
During the exposure time, photons in the incident light hits stripe pattern noise Nsp can be represented as
the sensing area of the sensor. Leveraging photoelectric con-
version, the sensor transforms photon to electrons. However, v
Nsp (n, m, λ) = Nsp
h
(n, m, λ) + Nsp (n, m, λ), (5)
due to the quantum property of light, the number of electrons
collected by sensor exists an ineluctable uncertainty, which h and N v denote horizontal and vertical stripe pat-
can be formulated as a Poisson distribution where Nsp sp
tern noise, respectively. The stripe pattern noise is caused
[L(m, n, λ) + Nsd (m, n, λ)] ∼ p(L(m, n, λ)), (2) by scanning camera design and not related to incident light,
and it is signal-independent noise. Following the assumption
where Nsd denotes the signal-dependent noise (i.e., shot of zero-mean Gaussian distribution in each row or column,
noise) and p(·) represents the Poisson distribution. Gener- we respectively formulate the horizontal and vertical stripe
ally, due to the limitation of light characteristic, shot noise pattern noise as
cannot be avoided by hyperspectral sensors. Overall, signal-
dependent noise is shot noise caused by photon-to-electrons Nsp
h
(m, λ) ∼ g(0, σh2 (λ)),
stage. v
(6)
Nsp (n, λ) ∼ g(0, σv2 (λ)),
123
International Journal of Computer Vision
σh (λ)
System gain k
Scales σ p (λ)
Scales σv (λ)
Parameter
h (m, λ)
transform on a bias frame, the highlighted vertical and horizontal pat-
terns in the centralized Fourier spectrum verify the existence of both
Nsp
v
horizontal and vertical stripe pattern noises, respectively
N = kNsd + Nsi ,
Photons-to-electrons
Photons-to-electrons
(8)
Electrons-to-voltage
= kNsd + N p + Nsp ,
Scanning system
Voltage-to-digit
where k, Nsd and Nsi represents system gain, signal-
dependent noise and signal-independent noise, respectively. Stage
parameter estimation method. According to Eqs. (2) (4) and Vertical stripe pattern noise Nsp
v
sented as
123
International Journal of Computer Vision
123
International Journal of Computer Vision
noisy HSI by the same factor to match the intensity of clean In this work, we propose a guided HSI denoising network,
HSI. which can effectively employ the prior of mean image of
Following these steps, we generate a realistic synthetic all spectral bands. Specifically, we design a guided dynamic
dataset with rich paired noisy and clean HSIs, which is bene- nonlocal attention module, where the attention map is cal-
ficial to train HSI denoising network and generalize it to the culated by the guidance information and related blocks are
evaluation on noisy HSI in the real world. dynamic for each spatial position.
To compare synthetic data by our noise model with real Our network has two branches, i.e., main denoising branch
data and synthetic data by previous noise model, e.g., and guidance information branch. Recent researches (Long
homoscedastic Gaussian noise model, heteroscedastic Gaus- et al. 2015; Chen et al. 2018; Xu et al. 2015) have shown the
sian noise model and complex noise model (Chen et al. 2017), effectiveness of CNN in image-to-image tasks, especially U-
we summarize the advantage and disadvantage of different Net (Ronneberger et al. 2015). Inspired by it, we employ a
data in Table 2. Collecting abundant high-quality real data is modified U-Net with 3D convolution as the main denoising
expensive and requires a amount of labor. In contrast, syn- branch to better exploit spectral correlation. The architec-
thesizing realistic paired data is convenient and inexpensive. ture of network is illustrated in Fig. 10. The main denoising
Synthetic data generated with our noise model is accurate, branch consists of 4 encoder stages and 4 corresponding
and can make the HSI denoising network trained on it per- decoder stages. At the end of each encoder stage, the feature
forms as well as that trained on real data. Besides, both real maps are downsampled to 1/2× scale with a 1 × 4 × 4 kernel
data and synthetic data generated with our noise model can size and 2 stride convolution. Before each decoder stage, the
be employed to benchmark the HSI denoising algorithms. feature maps are upsampled to 2× scale with a 1 × 2 × 2
kernel size and 2 stride transposed convolution. Skip con-
nections pass large-scale low-level feature maps from each
5 Guided HSI Denoising Network encoder stage to its corresponding decoder stage. We utilize
residual block as the fundamental block to build encoder and
In this section, we first describe the motivation of our method. decoder, which is conducted by two 3 × 3 × 3 convolutions
Then, we describe the overall network architecture of guided followed by LeakyReLU activation function and a 1 × 1 × 1
HSI denoising network. Finally, we introduce the guided convolution.
dynamic nonlocal attention (GDNA) module, which can For the guidance information branch, we employ a
adaptively guide the HSI denoising by the guidance infor- encoder with several residual blocks and downsampling lay-
mation of mean image of spectral bands. ers, and map the guidance information to desired spatial
resolution. Note that we only employ 2D convolution in
5.1 Motivation this branch as the mean image only has one channel and
no spectral correlation need to be exploited. Specifically, to
HSI has dozens to hundreds of spectral bands, and each band balance the performance and computational cost, we plug
often has random noises, including signal-dependent noise guided dynamic nonlocal attention module to 1/2× spatial
and signal-independent noise. For nature image denoising, resolution feature of main denoising branch in both encoder
researchers often capture several images and average them to and decoder.
obtain clean image (Abdelhamed et al. 2018). If we treat each We employ Charbonnier loss (Charbonnier et al. 1994) to
spectral band as a gray image and average them to obtain a optimize the proposed guided HSI denoising network, which
mean image, the signal-independent noise can be effectively can be formalized as
removed. Therefore, we can observe that the signal-to-noise
ratio of mean image of all spectral bands is higher than each
L(X̂ , X ) = ||X̂ − X ||2 + 2 , (13)
original band, as shown in Fig. 2. Besides, HSI often has
similar blocks in different spatial positions. On the basis of
this observation, traditional HSI denoising methods (Fu et al. where X̂ denotes the denoised HSI, and is a constant param-
2017; He et al. 2020) widely employ nonlocal self-similarity eter which is empirically set to 0.001.
prior, which calculates nonlocal correlation with dynamic
block groups. Inspired by that, recent deep learning methods 5.3 Guided Dynamic Nonlocal Attention
(Shi et al. 2021) design nonlocal attention network for HSI
denoising, however, which exploits nonlocal correlation in As mentioned in Sects. 2.2 and 2.3, self-guidance informa-
fixed neighborhoods. tion has been employed in several image restoration tasks
123
International Journal of Computer Vision
Real data
Synthetic data with previous noise model
Synthetic data with our noise model *
Note that synthetic data with our noise model is more accurate than that with previous model, and is realistic enough to benchmark HSI denoising
algorithm, even if it is not real data
Fig. 10 The overview of the proposed guided HSI denoising network a encoder to extract guidance information from mean image, and guided
architecture, which consists of main denoising branch and guidance dynamic nonlocal attention (GDNA) to guide HSI denoising. Note that
information branch. The main denoising branch employs U-Net as the we employ 3D convolution in the main denoising branch and 2D con-
basic architecture and residual block as fundamental block. We employ volutionn in the guidance branch, respectively
(a) (b)
Fig. 11 Guided dynamic nonlocal attention (GDNA)
and existing deep nonlocal methods employs a fixed number tion map is employed to aggregate feature of main denoising
of neighbor tokens for each query token. In this subsec- branch. The detailed structure of the GDNA module is shown
tion, we introduce a guided dynamic nonlocal attention in Fig. 11.
module, which can calculate dynamic nonlocal relationship
according to guidance information. It first employs the guid-
ance information to calculate initial correlation map. Then, 5.3.1 Initial Correlation Map
a dynamic gate mechanism is utilized to adaptively filter
the low-similarity tokens, and the remained high-correlation In our GDNA, we first employ two 1 × 1 convolution gφ and
tokens conduct the attention map. Finally, the dynamic atten- gψ to transform guidance information G ∈ RC g ×H ×W to two
123
International Journal of Computer Vision
Gφ = gφ (G), and Gψ = gψ (G). (14) Given the dynamic attention map, we attentively aggregate
feature of each band in main denoising branch, and the final
Then, we flatten Gφ and Gψ into feature vectors, and dot output can be expressed as
product them to calculate initial correlation map M ∈
R N ×N (N = H × W ). It can be represented as Fout,λ = Fin,λ + f out ( Ai, j f in (Fin,λ )), (19)
j∈ηi
M = θ (Gφ )T ⊗ θ (Gψ ), (15)
where f out is the output transform function with a 1 × 1
where θ is the flatten operation. convolution. As we separately calculate attention for each
spectral band, each spectral band can obtain different dif-
5.3.2 Dynamic Attention Map ferent dynamic attention map. Besides, it still maintains the
flexibility of 3D convolution network to process HSI with
To obtain the dynamic attention map, we design a dynamic different spectral bands.
gate mechanism, which can select high-similarity tokens with
a adaptive threshold. Let us consider Mi,: , i.e., the i-th row
of M, being correlations between the i-th token and the other 6 Experimental Results
tokens. We first average the Mi,: to obtain a initial threshold,
as it is the average importance of different tokens to the i-th In this section, we first introduce the settings in our exper-
token. To improve the adaptability, we further apply a token- iments, including implementation details and metrics for
specific linear transform to compute adaptive threshold Ti , quantitative evaluation. Then, we discuss the effect of dif-
which can be formalized as ferent noise model to synthesize noisy data. In addition, our
method is compared with several state-of-the-art methods on
)
τ1 (Fin,λ N our realistic HSI denoising dataset. Finally, we discuss the
Ti = Mi, j + τ2 (Fin,λ ) effect of different network modules.
N
j=1
(16)
α
N 6.1 Settings
= Mi, j + β,
N
j=1 6.1.1 Implementation Details
where Fin,λ is the λ-th band feature map transformed from To train our HSI denoising network, we crop overlapped
Fin,λ (Fin ∈ RΩ×C×H ×W ) by f in , and τ1 and τ2 are two 128 × 128 spatial regions from realistic dataset synthesized
convolution layer generating specific linear transform param- with our noise model and augment them by random flipping
eters α and β for each token, respectively. and/or rotation. Besides, we also train networks with our real
Then, we employ ReLU (Nair and Hinton 2010) as trun- HSI denoising dataset and employ the same augmentation
cation function, which is simply and differentiable. This method.
process removes the negative part and maintains the posi- Our implementation is based on PyTorch (Paszke et al.
tive part, and can be expressed as 2019). The models are trained with Adam optimizer (Kingma
and Ba 2014) (β1 = 0.9 and β2 = 0.999) for 200 epochs. The
Ai,: = r (Mi,: − Ti ), (17) initial learning rate and mini-batch size are set to 2 × 10−4
and 1, respectively.
where r is the ReLU function and A ∈ R N ×N is the dynamic
attention map. If Gi is highly correlated to Gi , Ai, j equals to 6.1.2 Evaluation Metrics
similarity weight, otherwise equals to zero.
Furthermore, we normalize non-zero weights of attention We employ four quantitative quality metrics to evaluate the
map by the softmax function. It can be represented as performance of all methods, including peak signal-to-noise
ratio (PSNR), structural similarity (SSIM) (Wang et al. 2004),
exp(Ai, j ) spectral angle mapping (SAM) (Kruse et al. 1993) and rela-
Ai, j = , j ∈ ηi , (18)
k∈ηi exp(Ai,k ) tive dimensionless global error in synthesis (ERGAS) (Wald
2000). PSNR and SSIM show the spatial accuracy, which
where ηi is the index group of high similarity tokens to the are calculate on each 2D spatial image and averaged over all
i-th token. spectral band. SAM and ERGAS show the spectral fidelity.
123
International Journal of Computer Vision
123
International Journal of Computer Vision
Table 5 Quantitative results of different methods on our real HSI denoising dataset
Metrics Methods
Noisy BM4D ITSReg LRTDTV NMoG LLRT NGmeet HSID-CNN QRNN3D 3D-ADNet Ours
PSNR 20.907 25.318 25.459 25.564 24.953 24.917 25.260 29.842 26.438 29.593 30.889
SSIM 0.3186 0.8156 0.8400 0.7859 0.7475 0.7399 0.7848 0.8825 0.7494 0.8641 0.9007
SAM 25.2992 6.3024 5.1425 6.4881 6.6469 7.4341 6.3595 5.1508 8.8176 6.1157 4.0924
ERGAS 60.758 35.078 34.618 34.429 37.163 36.698 35.037 22.399 48.289 23.009 20.516
Times(s) – 850 2717 482 662 1710 634 1 2 31 8
Fig. 13 Visual quality comparison on three typical outdoor scenes for noise removing in our real HSI denoising dataset. The denoising results and
the reference image are shown from left to right and from top to bottom
that our method can accurately formulate the noise in the HSI 2012), ITSReg (Xie et al. 2016a), LRTDTV (Wang et al.
and can generate the most realistic dataset for CNN training. 2017), NMoG (Chen et al. 2017), LLRT (Chang et al. 2017)
and NGmeet (He et al. 2019), and three learning-based
method, i.e., HSID-CNN (Yuan et al. 2018), QRNN3D (Wei
6.3 Evaluation of HSI Denoising Methods et al. 2021) and 3D-ADNet (Shi et al. 2021). We make great
effort to reproduce the best results for competitive methods
6.3.1 Compared Methods with the codes that are released publicly. Note that we train
all network with synthetic data generated by our noise model
We compare our method with four state-of-the-art HSI and test on our collected paired real dataset.
denoising methods on our captured real HSI dataset, includ-
ing six model-based methods, i.e., BM4D (Maggioni et al.
123
International Journal of Computer Vision
Fig. 14 Visual quality comparison on three typical indoor scenes for noise removing in our real HSI denoising dataset. The denoising results and
the reference image are shown from left to right and from top to bottom
Table 5 provides the averaged recovery results over all To visualize the experimental results, two representative
test images on our real HSI denoising dataset with indoor scenes are shown in Figs. 13 and 14. Compared with other
and outdoor scenes, to quantitatively compare our method methods, the recovered results by our method are consistently
with BM4D, ITSReg, LRTDTV, NMoG, LLRT, NGmeet, more accurate for all scenes. Specifically, our method can
HSID-CNN, QRNN3D and 3D-ADNet. The best results are produce the visually pleasant results with less artifacts and
highlighted in bold for each metric. It can be seen that sharp edges. It verifies that our method can provide higher
our method always has better performance than all com- spatial accuracy.
pared methods, which demonstrates the effectiveness of our
method. Moreover, our method outperforms existing deep 6.3.4 Spectral Fidelity
learning methods on all metrics. This reals the advantages of
mean image of spectral bands guidance and dynamic non- Figure 15 shows the recovered spectra of two points in the the
local correlation exploitation. Besides, we compare running selected images, which is indicated as orange in Fig. 13. We
time with other methods. It can be seen that the proposed can see that the spectra recovered by our method are much
method is comparable with HSID-CNN and QRNN3D and closer to the reference, which demonstrates that our method
is much faster than other methods, which verifies the effi- obtain higher spectral fidelity.
ciency of the proposed method.
123
International Journal of Computer Vision
123
International Journal of Computer Vision
Cao, X., Zhou, F., Xu, L., Meng, D., Xu, Z., & Paisley, J. (2018). for hyperspectral image mixed denoising. Signal Processing, 184,
Hyperspectral image classification with markov random fields and 108060.
a convolutional neural network. IEEE Trans Image Processing, He, K., Sun, J., & Tang, X. (2012). Guided image filtering. IEEE Trans
27(5), 2354–2367. Pattern Analysis and Machine Intelligence, 35(6), 1397–1409.
Chakrabarti, A., & Zickler, TE. (2011). Statistics of real-world hyper- He, W., Zhang, H., Zhang, L., & Shen, H. (2015). Total-variation-
spectral images. In: Proc. of Conference on Computer Vision and regularized low-rank matrix factorization for hyperspectral image
Pattern Recognition, pp. 193–200. restoration. IEEE Trans Geoscience and Remote Sensing, 54(1),
Chang, Y., Yan, L., & Zhong, S. (2017). Hyper-laplacian regularized 178–188.
unidirectional low-rank tensor recovery for multispectral image He, W., Yao, Q., Li, C., Yokoya, N., & Zhao, Q. (2019). Non-local meets
denoising. In: Proc. of Conference on Computer Vision and Pattern global: An integrated paradigm for hyperspectral denoising. In:
Recognition, pp. 4260–4268. Proc. of Conference on Computer Vision and Pattern Recognition,
Chang, Y., Yan, L., Fang, H., Zhong, S., & Liao, W. (2018). Hsi-denet: pp. 6868–6877.
Hyperspectral image restoration via convolutional neural network. He, W., Yao, Q., Li, C., Yokoya, N., Zhao, Q., Zhang, H., & Zhang,
IEEE Trans Geoscience and Remote Sensing, 57(2), 667–682. L. (2020). Non-local meets global: An integrated paradigm for
Chang, Y., Yan, L., Zhao, X. L., Fang, H., Zhang, Z., & Zhong, hyperspectral image restoration. IEEE Trans Pattern Analysis and
S. (2020). Weighted low-rank tensor recovery for hyperspectral Machine Intelligence Early Access.
image restoration. IEEE Trans Cybernetics, 50(11), 4558–4572. Healey, G. E., & Kondepudy, R. (1994). Radiometric ccd camera cal-
Charbonnier, P., Blanc-Feraud, L., Aubert, G., & Barlaud, M. (1994). ibration and noise estimation. IEEE Trans Pattern Analysis and
Two deterministic half-quadratic regularization algorithms for Machine Intelligence, 16(3), 267–276.
computed imaging. In: Proc. of International Conference on Image Holst, G.C. (1998). Ccd arrays, cameras, and displays.
Processing, 2, 168–172 Hui, TW., Loy, CC., & Tang, X. (2016). Depth map super-resolution by
Chen, C., Li, W., Tramel, E. W., Cui, M., Prasad, S., & Fowler, J. E. deep multi-scale guidance. In: Proc. of European Conference on
(2014). Spectral-spatial preprocessing using multihypothesis pre- Computer Vision, pp. 353–369.
diction for noise-robust hyperspectral image classification. IEEE Jiang, H., & Zheng, Y. (2019) Learning to see moving objects in the
Journal of Selected Topics in Applied Earth Observations and dark. In: Proc. of International Conference on Computer Vision,
Remote Sensing, 7(4), 1047–1059. pp. 7324–7333.
Chen, C., Chen, Q., Xu, J., & Koltun, V. (2018). Learning to see in Kawakami, R., Zhao, H., Tan, R. T., & Ikeuchi, K. (2013). Camera
the dark. In: Proc. of Conference on Computer Vision and Pattern spectral sensitivity and white balance estimation from sky images.
Recognition, pp. 3291–3300. International Journal of Computer Vision, 105(3), 187–204.
Chen, C., Chen, Q., Do, MN., & Koltun, V. (2019). Seeing motion in the Kingma, D.P., & Ba, J. (2014). Adam: A method for stochastic opti-
dark. In: Proc. of International Conference on Computer Vision, mization. arXiv preprint arXiv:1412.6980
pp. 3185–3194. Kruse, F. A., Lefkoff, A., Boardman, J., Heidebrecht, K., Shapiro, A.,
Chen, G., & Qian, S. E. (2010). Denoising of hyperspectral imagery Barloon, P., & Goetz, A. (1993). The spectral image processing
using principal component analysis and wavelet shrinkage. IEEE system (sips) interactive visualization and analysis of imaging
Trans Geoscience and Remote Sensing, 49(3), 973–980. spectrometer data. Remote Sensing of Environment, 44(2–3), 145–
Chen, Y., Cao, X., Zhao, Q., Meng, D., & Xu, Z. (2017). Denois- 163.
ing hyperspectral image with non-iid noise structure. IEEE Trans Kullback, S., & Leibler, R. A. (1951). On information and sufficiency.
Cybernetics, 48(3), 1054–1066. The Annals of Mathematical Statistics, 22(1), 79–86.
Dabov, K., Foi, A., Katkovnik, V., & Egiazarian, K. (2007). Image Kwon, H., & Nasrabadi, N. M. (2007). Kernel spectral matched filter for
denoising by sparse 3-d transform-domain collaborative filtering. hyperspectral imagery. International Journal of Computer Vision,
IEEE Trans Image Processing, 16(8), 2080–2095. 71(2), 127–141.
Dong, W., Li, G., Shi, G., Li, X., & Ma, Y. (2015). Low-rank tensor Li, Y., Huang, JB., Ahuja, N., & Yang, MH. (2016). Deep joint image
approximation with laplacian scale mixture modeling for multi- filtering. In: Proc. of European Conference on Computer Vision,
frame image denoising. In: Proc. of International Conference on pp. 154–169.
Computer Vision, pp. 442–449. Lin, B., Tao, X., & Lu, J. (2019). Hyperspectral image denoising via
Dong, W., Wang, H., Wu, F., Shi, G., & Li, X. (2019). Deep spatial- matrix factorization and deep prior regularization. IEEE Trans
spectral representation learning for hyperspectral image denoising. Image Processing, 29, 565–578.
IEEE Trans Computational Imaging, 5(4), 635–648. Liu, L., Jia, X., Liu, J., & Tian, Q. (2020). Joint demosaicing and denois-
Fu, Y., Lam, A., Sato, I., & Sato, Y. (2017). Adaptive spatial-spectral ing with self guidance. In: Proc. of Conference on Computer Vision
dictionary learning for hyperspectral image restoration. Interna- and Pattern Recognition, pp. 2240–2249.
tional Journal of Computer Vision, 122(2), 228–245. Liu, X., Bourennane, S., & Fossati, C. (2012). Denoising of hyperspec-
Fu, Y., Zheng, Y., Huang, H., Sato, I., & Sato, Y. (2018). Hyperspec- tral images using the parafac model and statistical performance
tral image super-resolution with a mosaic rgb image. IEEE Trans analysis. IEEE Trans Geoscience and Remote Sensing, 50(10),
Image Processing, 27(11), 5539–5552. 3717–3724.
Fu, Y., Zhang, T., Zheng, Y., Zhang, D., & Huang, H. (2019). Hyper- Long, J., Shelhamer, E., & Darrell, T. (2015). Fully convolutional
spectral image super-resolution with optimized rgb guidance. In: networks for semantic segmentation. In: Proc. of Conference on
Proc. of Conference on Computer Vision and Pattern Recognition, Computer Vision and Pattern Recognition, pp. 3431–3440.
pp. 11661–11670. Lu, G., & Fei, B. (2014). Medical hyperspectral imaging: A review.
Gu, S., Li, Y., Gool, LV., & Timofte, R. (2019). Self-guided network Journal of Biomedical Optics, 19(1), 010901.
for fast image denoising. In: Proc. of International Conference on Ma, C., Cao, X., Tong, X., Dai, Q., & Lin, S. (2014). Acquisition of high
Computer Vision, pp. 2511–2520. spatial and spectral resolution video with a hybrid camera system.
Guo, S., Liang, Z., & Zhang, L. (2021). Joint denoising and demosaick- International Journal of Computer Vision, 110(2), 141–155.
ing with green channel prior for real-world burst images. IEEE Maggioni, M., Katkovnik, V., Egiazarian, K., & Foi, A. (2012). Non-
Trans Image Processing, 30, 6930–6942. local transform-domain filter for volumetric data denoising and
He, C., Sun, L., Huang, W., Zhang, J., Zheng, Y., & Jeon, B. (2021). reconstruction. IEEE Trans Image Processing, 22(1), 119–133.
Tslrln: Tensor subspace low-rank learning with non-local prior
123
International Journal of Computer Vision
Miao, Y. C., Zhao, X. L., Fu, X., Wang, J. L., & Zheng, Y. B. Xie, Y., Qu, Y., Tao, D., Wu, W., Yuan, Q., & Zhang, W. (2016). Hyper-
(2021). Hyperspectral denoising using unsupervised disentangled spectral image restoration via iteratively regularized weighted
spatiospectral deep priors. IEEE Trans Geoscience and Remote schatten p-norm minimization. IEEE Trans Geoscience and
Sensing, 60, 1–16. Remote Sensing, 54(8), 4642–4659.
Monno, Y., Kiku, D., Tanaka, M., & Okutomi, M. (2015). Adaptive Xiong, F., Zhou, J., Zhao, Q., Lu, J., & Qian, Y. (2021). Mac-net: Model
residual interpolation for color image demosaicking. In: Proc. of aided nonlocal neural network for hyperspectral image denoising.
International Conference on Image Processing, pp. 3861–3865. IEEE Trans Geoscience and Remote Sensing.
Morgan, E. C., Lackner, M., Vogel, R. M., & Baise, L. G. (2011). Prob- Xu, L., Ren, J., Yan, Q., Liao, R., & Jia, J. (2015). Deep edge-aware
ability distributions for offshore wind speeds. Energy Conversion filters. In: Proc. of International Conference on Machine Learning,
and Management, 52(1), 15–26. pp. 1669–1678.
Nair, V., & Hinton, GE. (2010). Rectified linear units improve restricted Yasuma, F., Mitsunaga, T., Iso, D., & Nayar, S. K. (2010). Generalized
boltzmann machines. In: Proc. of International Conference on assorted pixel camera: Postcapture control of resolution, dynamic
Machine Learning. range and spectrum. IEEE Trans Image Processing, 19(9), 2241–
Ojha, L., Wilhelm, M. B., Murchie, S. L., McEwen, A. S., Wray, J. J., 2253.
Hanley, J., et al. (2015). Spectral evidence for hydrated salts in Yuan, Q., Zhang, L., & Shen, H. (2012). Hyperspectral image denoising
recurring slope lineae on Mars. Nature Geoscience, 8(11), 829– employing a spectral-spatial adaptive total variation model. IEEE
832. Trans Geoscience and Remote Sensing, 50(10), 3660–3677.
Paszke, A., Gross, S,. Massa, F., Lerer, A., Bradbury, J., Chanan, G., Yuan, Q., Zhang, Q., Li, J., Shen, H., & Zhang, L. (2018). Hyperspectral
Killeen, T., Lin, Z., Gimelshein, N., Antiga, L. et al. (2019). image denoising employing a spatial-spectral deep residual con-
Pytorch: An imperative style, high-performance deep learning volutional neural network. IEEE Trans Geoscience and Remote
library. arXiv preprint arXiv:1912.01703 Sensing, 57(2), 1205–1218.
Peng, Y., Meng, D., Xu, Z., Gao, C., Yang, Y., & Zhang, B. (2014). Zhang, H., He, W., Zhang, L., Shen, H., & Yuan, Q. (2013). Hyperspec-
Decomposable nonlocal tensor dictionary learning for multispec- tral image restoration using low-rank matrix recovery. IEEE Trans
tral image denoising. In: Proc. of Conference on Computer Vision Geoscience and Remote Sensing, 52(8), 4729–4743.
and Pattern Recognition, pp. 2949–2956. Zhang, L., Wei, W., Zhang, Y., Shen, C., Avd, Hengel, & Shi, Q. (2018).
Plotz, T., & Roth, S. (2017). Benchmarking denoising algorithms with Cluster sparsity field: An internal hyperspectral imagery prior for
real photographs. In: Proc. of Conference on Computer Vision and reconstruction. International Journal of Computer Vision, 126(8),
Pattern Recognition, pp. 1586–1595. 797–821.
Porter, WM., & Enmark, HT. (1987). A system overview of the airborne Zhang, T., Fu, Y., & Li, C. (2021). Hyperspectral image denoising with
visible/infrared imaging spectrometer (aviris). In: Proc. of Annual realistic data. In: Proc. of International Conference on Computer
Technical Symposium, pp. 22–31. Vision, pp. 2248–2257.
Ronneberger, O., Fischer, P., & Brox, T. (2015). U-net: Convolu- Zhao, B., Ulfarsson, M. O., Sveinsson, J. R., & Chanussot, J. (2022).
tional networks for biomedical image segmentation. In: Proc. Hyperspectral image denoising using spectral-spatial transform-
of International Conference on Medical image computing and based sparse and low-rank representations. IEEE Trans Geo-
computer-assisted intervention, pp. 234–241. science and Remote Sensing, 60, 1–25.
Schott, JR. (2007). Remote sensing: the image chain approach. Oxford Zheng, H., Ji, M., Wang, H., Liu, Y., & Fang, L. (2018). Crossnet: An
University Press on Demand. end-to-end reference-based super resolution network using cross-
Shi, Q., Tang, X., Yang, T., Liu, R., & Zhang, L. (2021). Hyperspectral scale warping. In: Proc. of European Conference on Computer
image denoising using a 3-d attention denoising network. IEEE Vision, pp. 88–104.
Trans Geoscience and Remote Sensing, 59(12), 10348–10363. Zhou, Y., Wu, G., Fu, Y., Li, K., & Liu, Y. (2021). Cross-mpi: Cross-
Wald, L. (2000). Quality of high resolution synthesised images: Is there scale stereo for image super-resolution using multiplane images.
a simple criterion? In: Proc. of Conference on Fusion of Earth In: Proc. of Conference on Computer Vision and Pattern Recogni-
Data, pp. 99–103. tion, pp. 14842–14851.
Wang, L., Zhang, S., & Huang, H. (2021). Adaptive dimension-
discriminative low-rank tensor recovery for computational hyper-
spectral imaging. International Journal of Computer Vision, Publisher’s Note Springer Nature remains neutral with regard to juris-
129(10), 2907–2926. dictional claims in published maps and institutional affiliations.
Wang, X., Yu, K., Dong, C., & Loy, CC. (2018). Recovering realistic
texture in image super-resolution by deep spatial feature trans- Springer Nature or its licensor (e.g. a society or other partner) holds
form. In: Proc. of Conference on Computer Vision and Pattern exclusive rights to this article under a publishing agreement with the
Recognition, pp. 606–615. author(s) or other rightsholder(s); author self-archiving of the accepted
Wang, Y., Peng, J., Zhao, Q., Leung, Y., Zhao, X. L., & Meng, D. manuscript version of this article is solely governed by the terms of such
(2017). Hyperspectral image restoration via total variation regu- publishing agreement and applicable law.
larized low-rank tensor decomposition. IEEE Journal of Selected
Topics in Applied Earth Observations and Remote Sensing, 11(4),
1227–1243.
Wang, Z., Bovik, A. C., Sheikh, H. R., & Simoncelli, E. P. (2004). Image
quality assessment: From error visibility to structural similarity.
IEEE Trans Image Processing, 13(4), 600–612.
Wei, K., Fu, Y., & Huang, H. (2021). 3-d quasi-recurrent neural network
for hyperspectral image denoising. IEEE Trans Neural Networks
and Learning Systems, 32(1), 363–375.
Xie, Q., Zhao, Q., Meng, D., Xu, Z., Gu, S., Zuo, W., & Zhang, L.
(2016a). Multispectral images denoising by intrinsic tensor spar-
sity regularization. In: Proc. of Conference on Computer Vision
and Pattern Recognition, pp. 1692–1700.
123