Improved L1/L2 minimization algorithm for segmental limited-angle CT reconstruction

Computed tomography (CT) technology is widely used in medical imaging, industrial non-destructive testing, and archaeological exploration. However, its application is often limited by issues such as high radiation doses or long scanning time. This paper utilizes the segmental limited-angle (SLA) sampling strategy to address these issues. However, SLA projections inherit limited-angle sampling properties, and shading artifacts still exist in the reconstructed images. To address this issue, we introduce the L1/L2 ratio of image gradients as a regularization term into SLA CT to construct a reconstruction model. The L1/L2 ratio is scale-invariant and can better approximate the L0 norm, shows great potential for improving image quality. To solve this model, we propose an improved L1/L2 minimization algorithm. First linearize the data fidelity term, and then use the Fast Fourier Transform (FFT) to accelerate the computation process. Finally, we employ the alternating direction method to obtain the reconstructed image. Numerical simulations and real CT data experiments demonstrate that the L1/L2 method outperforms other competing methods, and it can effectively preserve image structures and some details.