Scholarlink

<p>Hyperspectral image (HSI) denoising is a challenging task due to the presence of diverse and complex noise types. Such degradations significantly compromise image quality and restrict the performance of downstream applications. Although numerous HSI denoising methods have been proposed, achieving satisfactory restoration under severe noise conditions remains elusive. Unlike conventional images, HSIs not only exhibit nonlocal self-similarity within their spatial domain but also a high degree of spectral redundancy. This distinctive attribute offers an opportunity to exploit low-rank properties across both spectral and spatial dimensions. Building upon this premise, we propose a two-step denoising strategy based on the weighted Schatten <i>p</i>-norm. In the first step, weighted Schatten <i>p</i>-norm minimization (WSNM) is employed to capture spectral low-rank structure and suppress severe noise. In the second step, multi-band WSNM is applied to a low-rank matrix constructed by stacking nonlocally similar patches, which further removes residual noise while preserving texture details. Moreover, considering the non-convexity of the proposed model, we theoretically unify the two-step denoising process within a generalized soft-thresholding (GST) framework and analyze its computation complexity. Extensive experiments on both simulated and real datasets demonstrate that our method achieves superior quantitative and visual performance in severe noise removal compared to state-of-the-art approaches.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Hyperspectral image denoising via spectral-spatial weighted Schatten p-norm minimization

  • 期刊论文

  • 发表日期：

    2026年2月16日

查看全文
  • Dong Hu,
  • Tianyu Liu,
  • Yusen Tan,
  • Chuan Li,
  • Wu Chen,
  • Zhi Wang

摘要

Hyperspectral image (HSI) denoising is a challenging task due to the presence of diverse and complex noise types. Such degradations significantly compromise image quality and restrict the performance of downstream applications. Although numerous HSI denoising methods have been proposed, achieving satisfactory restoration under severe noise conditions remains elusive. Unlike conventional images, HSIs not only exhibit nonlocal self-similarity within their spatial domain but also a high degree of spectral redundancy. This distinctive attribute offers an opportunity to exploit low-rank properties across both spectral and spatial dimensions. Building upon this premise, we propose a two-step denoising strategy based on the weighted Schatten p-norm. In the first step, weighted Schatten p-norm minimization (WSNM) is employed to capture spectral low-rank structure and suppress severe noise. In the second step, multi-band WSNM is applied to a low-rank matrix constructed by stacking nonlocally similar patches, which further removes residual noise while preserving texture details. Moreover, considering the non-convexity of the proposed model, we theoretically unify the two-step denoising process within a generalized soft-thresholding (GST) framework and analyze its computation complexity. Extensive experiments on both simulated and real datasets demonstrate that our method achieves superior quantitative and visual performance in severe noise removal compared to state-of-the-art approaches.