Hyperspectral Mixed Noise Removal via Subspace Representation and Weighted Low-rank Tensor Regularization (2111.07044v1)

Abstract: Recently, the low-rank property of different components extracted from the image has been considered in man hyperspectral image denoising methods. However, these methods usually unfold the 3D tensor to 2D matrix or 1D vector to exploit the prior information, such as nonlocal spatial self-similarity (NSS) and global spectral correlation (GSC), which break the intrinsic structure correlation of hyperspectral image (HSI) and thus lead to poor restoration quality. In addition, most of them suffer from heavy computational burden issues due to the involvement of singular value decomposition operation on matrix and tensor in the original high-dimensionality space of HSI. We employ subspace representation and the weighted low-rank tensor regularization (SWLRTR) into the model to remove the mixed noise in the hyperspectral image. Specifically, to employ the GSC among spectral bands, the noisy HSI is projected into a low-dimensional subspace which simplified calculation. After that, a weighted low-rank tensor regularization term is introduced to characterize the priors in the reduced image subspace. Moreover, we design an algorithm based on alternating minimization to solve the nonconvex problem. Experiments on simulated and real datasets demonstrate that the SWLRTR method performs better than other hyperspectral denoising methods quantitatively and visually.

• The paper proposes SWLRTR, a novel model integrating subspace representation and weighted low-rank tensor regularization for effective hyperspectral image mixed noise removal.
• By projecting noisy images into a low-dimensional subspace, the method exploits global spectral correlation to simplify calculations and reduce computational complexity.
• Extensive experiments on real datasets demonstrate that SWLRTR quantitatively and qualitatively outperforms other state-of-the-art denoising algorithms.

The paper "Hyperspectral Mixed Noise Removal via Subspace Representation and Weighted Low-rank Tensor Regularization" presents an advanced approach for hyperspectral image (HSI) denoising, which addresses the limitations of traditional methods by preserving the intrinsic structural correlations of HSIs and reducing computational complexity.

Key Contributions

1. Novel Denoising Model:
   • The authors propose a model that integrates subspace representation with weighted low-rank tensor regularization (SWLRTR). This framework effectively removes mixed noise, including Gaussian, impulse, and other types common in HSI.
2. Subspace Representation:
   • By projecting noisy HSIs into a low-dimensional subspace, the method exploits the global spectral correlation (GSC) among spectral bands, facilitating a simpler calculation process.
   • The use of orthogonal matrices captures the high spectral correlation typical in HSIs, thus simplifying the problem to estimating subspace coefficients.
3. Weighted Low-Rank Tensor Regularization:
   • The approach uses a weighted tensor regularization to characterize image priors in a reduced subspace effectively. This step aids in preserving the tensor's intrinsic structure while mitigating computational complexity.
4. Optimization Algorithm:
   • An alternating minimization-based algorithm solves the nonconvex problem posed by the denoising model, ensuring optimal solution convergence.

Methodology

• The denoising process subdivides into spectral subspace projection and spatial low-rank tensor approximation, yielding a balance between computational efficiency and denoising performance.
• The method introduces blocks of patches from the HSI that are similar, reshaping them into 3D tensors, and employing a low-rank approximation to denoise the data.

Experimental Results

• The paper presents extensive experiments with both simulated and real datasets, including the Washington DC Mall, Pavia University, and Pavia Center datasets.
• Quantitative assessments using mean PSNR (MPSNR), mean SSIM (MSSIM), ERGAS, and mean spectral angle (MSA) show that SWLRTR performs favorably against other state-of-the-art denoising algorithms.
• Visual comparisons on datasets like KSC and Indian Pines demonstrate the method's efficacy in preserving both detail and structure in denoised images.

Sensitivity and Performance Analysis

• The paper discusses parameter sensitivity in terms of patch size, number of patches, subspace dimension, and regularization parameters. The stability across varying conditions suggests robust performance of the proposed method.
• The computational complexity analysis indicates that the method maintains a balance between denoising quality and processing time, making it pragmatic for large-scale hyperspectral data.

Conclusion

The proposed SWLRTR model significantly advances hyperspectral image denoising by integrating efficient tensor and subspace methodologies. It outperforms existing techniques in both quantitative and qualitative measures, making it a strong candidate for real-world HSI applications. Notably, the incorporation of iterative regularization further enhances the restoration accuracy, setting a new benchmark in the domain of hyperspectral denoising.