Hyperspectral imaging, providing abundant spatial and spectral informationsimultaneously, has attracted a lot of interest in recent years. Unfortunately,due to the hardware limitations, the hyperspectral image (HSI) is vulnerable tovarious degradations, such noises (random noise, HSI denoising), blurs(Gaussian and uniform blur, HSI deblurring), and down-sampled (both spectraland spatial downsample, HSI super-resolution). Previous HSI restoration methodsare designed for one specific task only. Besides, most of them start from the1-D vector or 2-D matrix models and cannot fully exploit the structurallyspectral-spatial correlation in 3-D HSI. To overcome these limitations, in thiswork, we propose a unified low-rank tensor recovery model for comprehensive HSIrestoration tasks, in which non-local similarity between spectral-spatial cubicand spectral correlation are simultaneously captured by 3-order tensors.Further, to improve the capability and flexibility, we formulate it as aweighted low-rank tensor recovery (WLRTR) model by treating the singular valuesdifferently, and study its analytical solution. We also consider the exclusivestripe noise in HSI as the gross error by extending WLRTR to robust principalcomponent analysis (WLRTR-RPCA). Extensive experiments demonstrate the proposedWLRTR models consistently outperform state-of-the-arts in typical low levelvision HSI tasks, including denoising, destriping, deblurring andsuper-resolution.
展开▼
