关键词:
低秩张量函数
高斯混合模型
低秩张量分解
噪声建模
摘要:
在高光谱图像(HSI)处理领域,一个核心问题是从各类退化现象中恢复HSI数据,诸如噪声污染及信息缺失等。然而,当前多数方法未能充分考量HSI的丰富先验信息及所含噪声,这在一定程度上限制了其实际应用效果。本文提出了一种新颖的HSI恢复方法,它结合了图像的内在结构和噪声特性,构建了一个基于低秩张量函数的模型。这个模型通过低秩张量分解来捕捉图像的全局结构,从而有效地利用图像的内在规律。同时,该模型还使用高斯混合模型来建模噪声的概率分布,这使得模型能够更好地适应不同类型和强度的噪声。在训练过程中,模型会综合考虑数据误差、图像平滑性约束以及负对数似然损失等多种因素。通过优化这些参数,模型能够进一步提高图像质量,实现清晰图像的恢复。在模拟和实际场景下的大量实验结果证明了所提方法的有效性以及其优于现有先进方法的优越性。In the field of hyperspectral image (HSI) processing, a core challenge is recovering HSI data from various degradation phenomena, such as noise contamination and missing information. However, most existing methods fail to fully account for the rich prior information in HSIs and the characteristics of the noise, which to some extent limits their practical effectiveness. This paper proposes a novel HSI recovery method that integrates the intrinsic structure of images and noise characteristics by constructing a model based on low-rank tensor functions. This model leverages low-rank tensor decomposition to capture the global structure of the image, effectively utilizing the inherent patterns within the data. Simultaneously, the model employs a Gaussian mixture model to characterize the probabilistic distribution of the noise, allowing it to better adapt to various types and levels of noise. During the training process, the model comprehensively considers factors such as data fidelity, image smoothness constraints, and negative log-likelihood loss. By optimizing these parameters, the model can further enhance image quality and achieve clear image recovery. Extensive experiments conducted in both simulated and real-world scenarios demonstrate the effectiveness of the proposed method and its superiority over state-of-the-art approaches.