IE-TV: Structural Information Enhanced 3-D Weighted Correlated Total Variation for Hyperspectral Image Denoising

作者：Ke Yang, Long Yu, Fan Li, Jia Chen, Zhaozhao Zeng, Jun Li, Antonio Plaza · 发表于：IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing · 年份：2025 · DOI：10.1109/jstars.2025.3649542 · 被引用次数：1 · 研究领域：Image and Signal Denoising Methods、Advanced Image Fusion Techniques、Remote-Sensing Image Classification

Hyperspectral image (HSI) denoising remains challenging due to the difficulty in recovering the complex structure in HSIs under the corruption of mixed-noise. Existing methods face two key limitations: 1) they ignore the underlying structural information, thus failing to preserve fine textures while smoothing the global structure, leading to the appearance of staircase artifacts or over-smoothing on spatial details; 2) the TV regularization overlooks the mixed differences of noise across three dimensions of HSI, thereby exacerbating the spectral over-smoothing and causing spectral distortion. To address these issues, we propose a structural information enhanced three-dimensional weighted correlated total variation (IE-TV) method for HSI denoising. We first construct a weighted CTV (WCTV) regularization term to adaptively adjust the smoothness constraints on different dimensions according to the noise characteristics. Furthermore, our IE-TV framework learns a structural information enhancement matrix at a more refined scale, which is then utilized to impose constraints on the gradient nuclear norm in WCTV. By solving this new optimization problem, IE-TV achieves a balance between the preservation of details (such as edges and textures) and the restoration of the global structure (such as spectral continuity). Experimental verifications on three HSI datasets show that IE-TV outperforms state-of-the-art methods in terms of both visual quality and quantitative indicators (PSNR, S...