中文摘要：

为更好地提高核磁共振图像重构质量，提出了一种基于熵约束字典学习和加权全变分的图像重构算法。首先对图像进行分块，基于熵约束构建新的字典学习模型，生成字典库；结合加权的各向同性与各向异性的全变分正则项构建图像重构模型，并采用Split—Bregman算法求解，最终得到重构图像。实验结果表明，该算法不仅能有效消除噪声，对噪声具有鲁棒性，又能保留图像边缘纹理信息，抑制阶梯效应。与现有的算法相比，该算法对图像重构有着更好的性能。

英文摘要：

In order to improve the quality of Magnetic Resonance Image reconstruction, this paper proposes a new reconstruction algorithm, which combines dictionary learning based on entropy-constraint with a weight total variation. Firstly, it blocks the image, constructs the proposed new model, and generates the dictionary library. Secondly, it constructs the image reconstruction model by combining a weight isotropic with anisotropic TV regularization. Finally, it obtains the reconstructed image using the Split-Bregman algorithm. Experimental results show that the proposed algorithm not only removes noise effectively and robust to noise ,but also preserves the texture and detail information better, greatly suppresses the staircase of the total variation. Comparing to the existing algorithms, the new algorithm has a better performance for image reconstruction.