中文摘要：

针对稀疏投影角度的CT图像重建问题，结合压缩感知理论，提出基于加权迭代支持检测的分块代数重建算法，以较少的投影角度重建出理想的CT图像。首先，针对传统的代数重建技术计算量大、收敛速度慢的问题，提出分块代数重建算法；其次，传统的最小总变差模型会引起图像过度平滑及纹理细节模糊等问题，对此提出一种最小加权总变差算法，即加权迭代支持检测算法，并建立加权迭代支持检测模型；最后，分块代数重建技术与加权迭代支持检测模型交替迭代，使重建结果趋于收敛。本文采用经典的Shepp-Logan体模及实际的脑部CT切片进行重建，以均方根误差作为重建图像的质量评判标准，并与其他重建算法的重建结果进行对比。在经过一定次数的迭代后，基于本文算法的重建图像更贴近原始图像，而且比其他算法更早收敛。实验结果表明，本文算法在重建质量及收敛速度上都优于其他对比算法。

英文摘要：

Combined with compressed sensing theory, block algebraic reconstruction technique based on reweighted iterative support detection （Block-ART-RISD） was proposed for resolving the problem of CT image reconstruction of sparse projection angles, reconstructing satisfactory CT images with less sparse projection angles. A Block-ART was firstly proposed to overcome the large calculation amount and slow convergence speed of the traditional ART. A reweighted total variation minimization algorithm, also named reweighted iterative support detection （RISD） algorithm, was proposed, because the traditional total variation minimization model always results in reconstructed images of excessive smoothing and fuzzy texture details. And a RISD model was established. The Block-ART alternately iterated with RISD model to make the reconstructed results tend to convergence. The classical Shepp-Logan phantom was applied to reconstruct a real brain CT slice. The root mean square error was used as the evaluation criteria for the reconstructed image quality. The reconstructed image was also compared with the image reconstructed by other methods. After a certain number of iterations, the image reconstructed by proposed method was closer to the original image, with a faster convergence speed. The experiment results show the proposed method was better than other methods in reconstructed quality and convergence speed.