为了减轻X射线辐射对病患的危害，CT扫描设备的设计需要考虑降低辐射剂量的需求。源于压缩感知理论的字典 学习重建算法可用于求解不完备投影数据的CT图像重建问题，然而该算法的正则约束项无法有效区分噪声和低对比度信 息，重建图像容易丢失软组织边缘细节信息。本文提出一种加权字典学习重建算法，基于每次迭代的结果，利用字典稀疏 表示的残差设计正则约束项的权重因子，使算法在迭代过程中对图像不同区域施加不同程度的平滑效应，从而在平滑噪声 的同时保留低对比度信息。实验结果表明，提出的改进算法有效的保留了软组织边缘信息，与原算法相比，明显提高了重 建图像的质量。

To prevent the patients from the overdose of X-ray radiation, the radiation dose should be reduced in the design of the CT scanning device. The dictionary learning reconstruction algorithm, which is derived from the compressed sensing theory, is able to make high quality recovery from the under-sampled scanning data. However, the regularization term of this algorithm is not able to distinguish the noise and the low-contrast information, tending to lose the soft tissue edge details. This article proposed a weighted dictionary learning method, which calculated the weight factors of the regularization term based on the residuals of the image patches by subtracting the dictionary sparse representation after each iterative step. The weight factors helped preserve structural information of the image and smooth the noise during the iterative process by changing the smooth effects upon different regions of the image. The corresponding experiments proves that the proposed algorithm preserves the soft tissue edge details efficiently. The quality of the reconstruction image is improved compared with the existing dictionary learning method.