Scholarlink

<p>Sparse angle CT, as an advanced CT technology, its image reconstruction algorithm is currently a research hotspot. Addressing the issues of noise and artifacts in low contrast areas of images in existing group sparse regularized sparse angle CT reconstruction algorithms; this paper proposes a two regularization CT reconstruction model (PLS-GSR-RoG) that combines group sparsity (GSR) and Relativity-of-Gaussian (RoG) to complement each other. The GSR term fully considers the local sparsity and non local self similarity of the image, enabling the algorithm to effectively capture sharp edges and fine textures; The RoG term can recognize more similar directional gradients, and by globally optimizing the features of gradient domain images at different scales, the algorithm can effectively smooth out noise and artifacts in low contrast areas while preserving image structural information; Meanwhile, as the number of iterations increases, in order to avoid the RoG regularization term causing the image to be too smooth, the iteration process is divided into two stages. In the first stage, a GSR and RoG dual regularization term model is used, and in the second stage, only the GSR regularization term is used to reconstruct more image details. This paper verifies the advantages of the two-stage PLS GSR RoG through experiments. In the FORBILD head phantom and thoracic image reconstruction experiments at different projection angles, the experimental comparison with several commonly used CT image reconstruction algorithms (SART-TV, SART-RTV, SART-GSR, and SART-GSR-WIGF) shows that all test images achieved the best overall performance using 32 and 64 projection angles. Representative results show that the PSNR of the reconstructed image reached 47.18&#xa0;dB, and the FSIM reached 0.9996. In addition, the model ablation experiments demonstrated the effectiveness of combining GSR with RoG in a staged iterative manner. To further validate the approach, pelvic images were included for testing, achieving a PSNR of 48.03&#xa0;dB and an FSIM of 0.9988 in the reconstructed images.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Two-Stage Sparse Angle CT Reconstruction Combining Group Sparsity and Relativity-of-Gaussian

  • 期刊论文

  • 发表日期：

    2025年12月10日

查看全文
  • Yan Ma,
  • Yanping Bai,
  • Ting Xu,
  • Rong Cheng

摘要

Sparse angle CT, as an advanced CT technology, its image reconstruction algorithm is currently a research hotspot. Addressing the issues of noise and artifacts in low contrast areas of images in existing group sparse regularized sparse angle CT reconstruction algorithms; this paper proposes a two regularization CT reconstruction model (PLS-GSR-RoG) that combines group sparsity (GSR) and Relativity-of-Gaussian (RoG) to complement each other. The GSR term fully considers the local sparsity and non local self similarity of the image, enabling the algorithm to effectively capture sharp edges and fine textures; The RoG term can recognize more similar directional gradients, and by globally optimizing the features of gradient domain images at different scales, the algorithm can effectively smooth out noise and artifacts in low contrast areas while preserving image structural information; Meanwhile, as the number of iterations increases, in order to avoid the RoG regularization term causing the image to be too smooth, the iteration process is divided into two stages. In the first stage, a GSR and RoG dual regularization term model is used, and in the second stage, only the GSR regularization term is used to reconstruct more image details. This paper verifies the advantages of the two-stage PLS GSR RoG through experiments. In the FORBILD head phantom and thoracic image reconstruction experiments at different projection angles, the experimental comparison with several commonly used CT image reconstruction algorithms (SART-TV, SART-RTV, SART-GSR, and SART-GSR-WIGF) shows that all test images achieved the best overall performance using 32 and 64 projection angles. Representative results show that the PSNR of the reconstructed image reached 47.18 dB, and the FSIM reached 0.9996. In addition, the model ablation experiments demonstrated the effectiveness of combining GSR with RoG in a staged iterative manner. To further validate the approach, pelvic images were included for testing, achieving a PSNR of 48.03 dB and an FSIM of 0.9988 in the reconstructed images.