Real Noise Decoupling for Hyperspectral Image Denoising

TL;DR

The paper tackles real-world hyperspectral image denoising by decoupling complex noise into an explicitly modeled part $N_e$ and an implicitly modeled part $N_i$, enabling targeted pre-training and refinement. It introduces EMNet, pre-trained on synthetic explicit noise, and IMNet, guided by high-frequency wavelet features to remove implicit noise, connected via a three-stage learning strategy that mitigates error accumulation. Across MEHSI and RealHSI datasets, the approach achieves state-of-the-art PSNR, SSIM, and SAM improvements, demonstrating robustness to diverse, real-world noise patterns and better spectral fidelity. This work offers a practical, generalizable framework for real HSI denoising with clear pathways for leveraging physical noise models and wavelet-guided feature extraction.

Abstract

Hyperspectral image (HSI) denoising is a crucial step in enhancing the quality of HSIs. Noise modeling methods can fit noise distributions to generate synthetic HSIs to train denoising networks. However, the noise in captured HSIs is usually complex and difficult to model accurately, which significantly limits the effectiveness of these approaches. In this paper, we propose a multi-stage noise-decoupling framework that decomposes complex noise into explicitly modeled and implicitly modeled components. This decoupling reduces the complexity of noise and enhances the learnability of HSI denoising methods when applied to real paired data. Specifically, for explicitly modeled noise, we utilize an existing noise model to generate paired data for pre-training a denoising network, equipping it with prior knowledge to handle the explicitly modeled noise effectively. For implicitly modeled noise, we introduce a high-frequency wavelet guided network. Leveraging the prior knowledge from the pre-trained module, this network adaptively extracts high-frequency features to target and remove the implicitly modeled noise from real paired HSIs. Furthermore, to effectively eliminate all noise components and mitigate error accumulation across stages, a multi-stage learning strategy, comprising separate pre-training and joint fine-tuning, is employed to optimize the entire framework. Extensive experiments on public and our captured datasets demonstrate that our proposed framework outperforms state-of-the-art methods, effectively handling complex real-world noise and significantly enhancing HSI quality.

Figures (6)

• Figure 1: The significant differences between real noisy HSIs and synthetic noisy HSIs generated by the noise model zhang2022guided.
• Figure 1: Summary of existing datasets and our multi-exposure dataset.
• Figure 2: The overall multi-stage noise-decoupling framework. Stage 1: EMNet removes explicitly modeled noise based on the noise model, guiding the IMNet (Stage 2) to adaptively eliminate implicitly modeled noise. Stage 3: Entire fine-tuning for mitigating noise accumulation errors across stages and effectively removing real-world noise.
• Figure 3: Example scenes in MEHSI dataset.
• Figure 4: Visual quality comparisons of sample scene on MEHSI (up) and RealHSI (bottom) datasets with the spectral band in 510 $nm$, and 550 $nm$, respectively. The 'corr' of spectral density curves in the color box represents the correlation coefficient.