Magnetic resonance images are accompanied by random Rician noise due to the influence of uncertain factors in the process of imaging, storage, which brings a lot of inconvenience to the subsequent processing of the im...
详细信息
ISBN:
(纸本)9783031189098;9783031189104
Magnetic resonance images are accompanied by random Rician noise due to the influence of uncertain factors in the process of imaging, storage, which brings a lot of inconvenience to the subsequent processing of the image and clinical diagnosis. This paper proposes an improved multipath matching pursuit algorithm based on learning Gabor pattern dictionary atom for image reconstruction and denoising. Firstly, Gabor wavelet transform based on neurophysiological constraints is used to generate dictionary atoms that match the local features of the image;Then this paper introduces adaptive differential evolution algorithm optimization to the process of solving multiple candidate atoms matching the local image features in each iteration of the multipathmatchingpursuit. It combines the advantages of adaptive differential evolution and multipath matching pursuit algorithm, not only avoids the genetic falling into the local optimal defect, but also obtains the best matching parameters with higher accuracy, and effectively reduces the computational complexity of the multipathmatchingpursuit. In the reconstruction experiment of the simulated MR images, compared with state-of-the-art denoising algorithms, our algorithm not only shows better denoising performance, but also retains more detailed information, and the running time is reduced nearly 50% than multipathmatchingpursuit;which fully shows the clinical application value.
To increase the possibility of choosing the true support of a sparse signal, multipathmatchingpursuit (MMP) algorithm generates multiple promising candidates of the support set by tree-searching structure. This Lett...
详细信息
To increase the possibility of choosing the true support of a sparse signal, multipathmatchingpursuit (MMP) algorithm generates multiple promising candidates of the support set by tree-searching structure. This Letter proposes the restricted isometry property-based conditions under which at least one candidate of MMP contains the true support set in cases of $l_algorithm$l2 and $l_{\infty }$l infinity bounded noises. Comparison with the existing result is also presented.
暂无评论