Patch-wise non-local low-rank for few-view multi-energy CT reconstruction

With the development of detection technology like the newest photon counting detector, multi-energy CT has been a hot topic in recent years. Many algorithms have been proposed to exploit the similarity and correlation within multi-energy images, some of them is based on low-rank approximation, such as newly proposed PRISM method by Hao Gao et.al. However, conventional low-rank methods assume that the image itself is low-rank, or there is strong correlation between its columns or rows. This assumption does not hold for most images in clinical practice. Very recently, a patch-wise non-local low-rank method named LASSC (Low-rank Approximation towards Simultaneous Sparse Coding) has been proposed for image denoising and restoration. It assumes that the group of similar patches should be low-rank, which is valid for most natural images. In this work, we adopt LASSC as the regularization term for few-view CT reconstruction, and achieved well-preserved fine structures taking projections at only 16 views in numerical simulation. After incorporating a TV transform, our method can be extended to multi-energy CT reconstruction, and this work is still underway.