Nonlocal means exploits the spatial correlation in an image

and preserves the structure information effectively.Combining the nonlocal patch regularization with TV regularization

we propose a nonlocal patch regularized image restoration model.The improved structure tensor matrix can be used to achieve a data-adaptive weigh function

which can then adjust the similarity match process based on the local structure of a pixel.A more simple and effective algorithm -Split Bregman

is used to solve the model iteratively.Compared with other regularization models

our method performs better in improving the quality of restoration image and the efficiency of the algorithm.