The current methods of motion artifact suppression in magnetic resonance imaging(MRI) are generally based on the K-space data.In this paper

an image post-processing method is proposed

which can directly apply to the MR images and correct the motion artifact.We designed a model based on non-local means total variational (TV) method

which consists of the non-local regularization term and patch similarity fidelity term.The regularization term can keep the image structure while correcting the motion artifact.Introduce the anisotropic diffusion structure tensor to the patch similarity fidelity term as its weight function

which can implement different diffusions process in different regions and can remain the details of the image while removing the motion artifact.The numerical scheme uses the split Bregman method.Experiments showed that the motion artifact was effectively reduced and the valuable details of images were remained and the arithmetic speed was improved.