Local low-rank matrix recovery for hyperspectral image denoising with ℓ0 gradient constraint
Introduction
Hyperspectral images are acquired by imaging spectrometers over hundreds of bands and have rich spectral information compared with that of natural images. Unfortunately, the visual quality of HSI is inevitably damaged by different types of noise during the acquisition and transmission procedure [1], [2], which severely hinders the precision of the subsequent editing and rendering tasks, e.g., classification [3], segmentation [4], and matching [5]. Therefore, HSI denoising has become an essential preprocessing step for subsequent exploitation [6], [7], [8].
Hyperspectral images restoration is a well-studied problem [9], [10], [11], [12]. Traditional 2-D denoising methods [13] can be directly applied to restore HSI by regarding each band as one independent gray image, e.g. [14], [15]. However, the performance of the bandwise methods is limited since the strong correlation among adjacent bands are ignored. Inspired by the global spectral correlation of the clean HSI, low-rank approximation methods [16], [17], [18] have been widely adopted to HSI noise removal and achieved significant success. For example, Zhang et al. [19] employ low-rank matrix recovery (LRMR) for HSI denoising with remarkable improvement. To handle defective pixels in some channels, sparse penalty is introduced into a low-rank framework for HSI reconstruction in [20]. These methods ignore the spatial prior information of HSI. To explore the spatial structure, many traditional spatial regularizers, such as nonlocal similarity [21], total variation [22], anisotropic diffusion [23], are embedded into the low-rank based method. Motivated by the outstanding performance of the total variation (TV) regularization in image processing tasks, three-dimensional TV [24] and spatio-spectral TV (SSTV) [25] have been developed and involved in the low-rank framework for HSI mixed-noise removal, achieving state-of-the-art results.
For real-world HSIs, besides the correlations exhibited by the adjacent spectral channels, it is well known that the nearby pixels are also highly correlated. That is to say, the nonlocal similarity would effectively enhance the low-rank property of the hyperspectral imagery. Zhang et al. [19] first divide the HSI into patches and restore them sequentially. Based on the patch-wise idea, a noise-adjusted iterative mechanism is adopted in [27] to make LRMR adaptive to the different corruption levels in different bands. Lu et al. [28] introduce a novel spatial-adaptive similar pixel searching strategy to group the similar pixels. To exploit both the local similarity and the global spatial-spectral structure, He et al. [35] propose to combine the global SSTV regularization model with local patch-based rank-constrained robust PCA (RPCA) model for HSI noise removal.
Theoretically, ℓ0 norm measures the sparsity of gradients. Furthermore, ℓ0 gradient minimization [29], [30] can be applied to images to produce sparse solutions, e.g. as a powerful sparsity measure method. Recently, the ℓ0 gradient minimization is used to the task of image restoration [31], [32]. Due to the discrete nature and nonconvexity of the ℓ0 norm, minimizing ℓ0 gradient is NP-hard and intractable. As such, a number of convex relaxation strategies have been developed [32], [33]. Inspired by the strong ability of sparsity-control of ℓ0 gradient minimization, we combine the ℓ0 gradient with the low-rank matrix factorization and propose a unified mixed noise removal framework named ℓ0 gradient constrained local low-rank matrix recovery (ℓ0-LLRMR) for HSI denoising.
We adopt the patch-based restoration schema to enhance the low-rank property and restrict the ℓ0 gradient value of the restored image to preserve the spatial piecewise smooth information. Moreover, users can directly impose a desired sparsity of the restored image by α, i.e., the ℓ0 gradient no more than a user-given parameter α. Here, we use the percentage of the total pixels of the observed image to determine α. In our model, the parameter α has intuitive meaning that denotes a maximum ℓ0 gradient value of the output image. Apart from the non-convexity of the ℓ0 gradient, the proposed model is also a constrained optimization problem. To overcome this remedy, we develop an iterative schema based on the augmented Lagrange multiplier (ALM) algorithm [34] and derive closed-form solutions of the subproblems for calculating an approximate solution of the nonconvex optimization problem. Moreover, we present the convergence tendency of the objective function.
Section snippets
Patch based LRMR model
Observed HSI data, which is denoted by is always corrupted by mixed noise. The contaminated HSI can be formulated aswhere and represent the clean HSI and the mixed noise term, respectively. According to the real situations, we divide the mixed noise term into two types, i.e., the Gaussian noise and the sparse noise which represents the mixture of impulse noise, deadlines, and stripes. More specifically, the HSI degradation model can be expressed as
For a patch
Formulation of ℓ0 gradient constraint
Considering that ℓ0 norm directly measures sparsity of a vector, and ℓ0 gradient minimization has strong ability of edge preservation, this paper defines a HSI denoising technique which combines the ℓ0 gradient for the constraint of spatial smoothness and the local low-rank matrix recovery to explore the spectral low-rank property. Accordingly, the proposed ℓ0-LLRMR method is formulated as follows:where α is a
Performance verification
In this section, to evaluate the performance of our proposed method, we conduct experiments on simulated and real HSI datasets. To thoroughly assess the proposed algorithm, three representative denoising approaches are selected for comparison, i.e., TV-regularized low-rank matrix factorization (LRTV) [26], SSTV regularized local low-rank matrix recovery (LLRGTV) [35], and subspace-based nonlocal low-rank and sparse factorization (SNLRSF) [38]. For competitors, the parameters are optimally
Conclusion
This paper has proposed a novel low-rank matrix recovery and ℓ0 gradient constraint method for HSI mixed-noise reduction. In this approach, firstly, exploits the patch-based low-rank property of HSI to separate the clean signal from the sparse and Gaussian noise. Meanwhile, a global ℓ0 gradient constraint is utilized to recover the global spatial-spectral piecewise smoothness of the HSI. Specifically, we adopt parameter α to directly constraint the image smoothness of the output image. In
Declaration of Competing Interest
None.
Acknowledgments
This work was supported in part by the National Key R&D Program of China under Grant 2018YFB1305200, the National Natural Science Foundation of China under Grant 61602413 and Grant U1509207, and the Natural Science Foundation of Zhejiang under Grant LY19F030016.
References (38)
- et al.
Hyperspectral image denoising with superpixel segmentation and low-rank representation
Inf. Sci.
(2017) - et al.
Denoising point sets via ℓ0 minimization
Comput. Aided Geom. Des.
(2015) - et al.
Feature-preserving filtering with ℓ0 gradient minimization
Comput. Graph.
(2014) - et al.
Flexible FTIR spectral imaging enhancement for industrial robot infrared vision sensing
IEEE Trans. Ind. Inform.
(2020) - et al.
RISIR: Rapid infrared spectral imaging restoration model for industrial material detection in intelligent video systems
IEEE Trans. Ind. Inform.
(2019) - et al.
Spectral-spatial sparse subspace clustering for hyperspectral remote sensing images
IEEE Trans. Geosci. Remote Sens.
(2016) - et al.
An active learning framework for hyperspectral image classification using hierarchical segmentation
IEEE J. Sel. Top. Appl. Earth Observ. Remote Sens.
(2016) - et al.
Robust feature matching for remote sensing image registration via locally linear transforming
IEEE Trans. Geosci. Remote Sens.
(2015) - et al.
Hyperspectral image classification with global-local discriminant analysis and spatial-spectral context
IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens.
(2018) - et al.
Sparse unmixing with dictionary pruning for hyperspectral change detection
IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens.
(2017)
Hyperspectral image super-resolution via non-local sparse tensor factorization
Proc. IEEE Comput. Vis. Pattern Recognit.
Adaptive spatial-spectral dictionary learning for hyperspectral image restoration
Int. J. Comput. Vis.
HSI-Denet: hyperspectral image restoration via convolutional neural network
IEEE Trans. Geosci. Remote Sens.
Weighted joint sparse representation for removing mixed noise in image
IEEE T. Cybern.
Efficient blind signal reconstruction with wavelet transforms regularization for educational robot infrared vision sensing
IEEE-ASME Trans. Mechatron.
Nonlocal image restoration with bilateral variance estimation: a low-rank approach
IEEE Trans. Image Process.
Weighted Nuclear Norm Minimization with Application to Image Denoising
Proc. IEEE Comput. Vis. Pattern Recog.
Robust Principal Component Analysis: Exact Recovery of Corrupted Low-rank Matrices via Convex Optimization
Proc. NIPS
Fast hyperspectral image denoising and inpainting based on low-rank and sparse representations
IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens.
Cited by (4)
Hyperspectral image denoising using constraint smooth rank approximation and weighted enhance 3DTV
2022, DisplaysCitation Excerpt :In recent years, the low-rank (LR) prior information and band correlation of HSI have been widely studied. Framework of LR has good performance in HSI denoising and compressive perception [7-10]. Zhang et al. [11] proposed a method based on low-rank matrix restoration (LRMR), which expanded the three-dimensional HSI along the spectral dimension to form a two-dimensional HSI matrix, and successfully recovered a clean image from the damaged HSI.
Fusion of Hyperspectral-Multispectral images joining Spatial-Spectral Dual-Dictionary and structured sparse Low-rank representation
2021, International Journal of Applied Earth Observation and GeoinformationCitation Excerpt :The HSIs possess abundant redundancy, that is, spectral global correlation and spatial nonlocal similarity, which have significantly improved HSI restoration methods (Chen et al., 2020; Xue et al., 2019). Low-rank representation (LRR) projects high-dimensional signals into a lower-dimensional subspace, then uses a sparse linear combination of dictionary atoms to recover the underlying structure hidden in the original data (Pan et al., 2018; Yang et al., 2020). Since there are the spectral correlations along with successive spectral bands and the spatial correlations among spatial nonlocal similarity patches in HSIs, the design of a rational LRR constraint to represent such correlations is key to regularizing the ill-posed problem of the HSI-MSI fusion task.
A PARAMETERIZED THREE-OPERATOR SPLITTING ALGORITHM FOR NON-CONVEX MINIMIZATION PROBLEMS WITH APPLICATIONS
2024, Journal of Nonlinear and Variational AnalysisResearch on the denoising algorithm for hyperspectral images based on tensor decomposition and full variational constraints
2024, Applied Mathematics and Nonlinear Sciences