A new super resolution and deblurring algorithm for Magnetic Resonance images based on sparse representation and dictionary learning

Abstract

Magnetic Resonance Imaging (MRI) provides a non-invasive manner to aid clinical diagnosis, while its limitation is the slow scanning speed. Recently, due to the high costs of health care and taking account of patient comfort, some methods such as Parallel MRI (pMRI) and compressed sensing MRI have been developed to reduce the MR scanning duration under the sampling process. It is almost unavoidable to accept some doses of X-rays in computed tomography (CT scans). If one could find a more efficient way to represent the required visual information, the tasks of image processing and medical imaging would become easier and less troublesome. In this paper, first, we used pMRI on complex double data of brain magnetic resonance image. pMRI significantly reduces the number of measurements in the Fourier domain because each coil only acquires a small fraction of the whole measurements. It is important to reconstruct the original MR image efficiently and precisely for better diagnosis. In this research, we proposed a new super resolution and deblurring algorithm with dictionary learning, based on assuming a local Sparse-Land model on image patches, serving as regularization, then we validated the proposed method by using another one called the adaptive selection of sub dictionaries- adaptive reweighted sparsity regularization. Visual comparison and significant difference in psnr calculation (0.8111db) and time complexity showed that the proposed method had much better results.