Multispectral Image Noise Removal With Adaptive Loss and Multiple Image Priors Model

Multispectral image (MSI) denoising is a crucial pre-processing step for various subsequent image processing tasks, including classification, recognition, and unmixing. This paper proposes a novel image denoising model that integrates both noise modeling and image prior knowledge modeling. Specifically, to account for the complexity and non-uniformity of noise, a non-independent identically distributed mixture of Gaussian model is employed for noise modeling, and a weighted loss function is obtained. The weights used in the loss function are adaptively learned from noisy MSI and employed to adjust the denoising strength of each pixel. Additionally, the model leverages the prior knowledge of the image by utilizing a nonlocal low-rank matrix model that captures the spatial-spectral correlation and nonlocal spatial similarity priors of the image. Moreover, our model adopts the weighted spatial-spectral TV model to encode the local smoothness prior of the image. Both prior models are translated into regularization terms in the denoising model. The efficacy of the proposed method is demonstrated through both simulated and real image experiments.