双正则化参数法超分辨率重建核磁共振图像

针对基于稀疏编码的超分辨率算法噪点、伪影较多的问题,提出一种双正则化参数核磁共振图像超分算法。该算法引入在线字典学习方法,以训练正则化参数λt分开训练生成精确的超完备字典对,并调整重建正则化参数λr,得到最佳的稀疏系数用于恢复目标高分图像。实验结果表明:改进算法比双字典学习超分法的目标图像峰值信噪比和结构相似性平均值分别提高了1.30 d B和0.023,有效地抑制了噪点和边缘伪影,较大幅度地提升了核磁共振图像的超分效果。

MRI Super-resolution Reconstruction via Double Regularization Parameters

To solve the problem that there are severe artifacts and noise in the images recovered up-scaled by the super-resolution (SR) algorithms based on sparse coding, an improved SR algorithm for MRI images based on double regularization parameters is proposed in this paper.The proposed algorithm introduces Online Dictionary Learning method and train to acquire the over-complete dictionary pair with the training regularization parameter λt.Then the reconstruction regularization parameter λr is tuned to solve the best reconstruction sparse coefficient to recover the target high-resolution image.In the experiments, the average PSNR and SSIM of the reconstructed images with the proposed algorithm is 1.30 dB and 0.023 higher than the Couple Dictionary Learning SR algorithm.The SR performance is raised considerably to eliminate the noise and artifacts effectively.