【摘要】 为提高磁共振图像的重构质量,提出一种基于非参数贝叶斯分类字典学习的重建方法。通过差分变换,在梯度域中利用无限高斯混合模型将图像块自动聚类,对具有相似结构的图像块进行分类训练字典。采用非参数贝叶斯字典学习方法训练字典,克服传统字典学习对参数选择的依赖性。实验结果表明,与目前几种典型的磁共振图像重建方法相比,该方法的峰值信噪比平均提高2.9dB;在同一噪声水平下,该方法抗噪性能更强,重构质量更优。更多还原

【Abstract】 To improve the reconstruction quality of magnetic resonance images,a reconstruction method based on nonparametric Bayesian group dictionary learning was proposed.Through differential transformation,the image patches were automatically clustered through infinite Gaussian mixture model in the gradient domain.Group dictionaries for image patches with the similar structure were trained.A nonparametric Bayesian dictionary learning method was used to train the dictionary to alleviate the dependence of traditional dictionary learning on parameter selection.Experimental results show that the peak signal-to-noise ratio of the proposed method is improved by 2.9 dB on average compared with that of the typical methods,and the anti-noise performance and the reconstruction quality of the proposed method are better under the same noise level.更多还原