Super Resolution Reconstruction of Brain MR Image Based on Convolution Sparse Network

In order to recover high resolution image from their corresponding low-resolution counterparts for MR Images, this paper has proposed a super resolution reconstruction method to recover the low-resolution MR images based on convolution neural network. Based on the proposed network, the convolution operation and non-linear mapping are employed to adapt MR images naturally and leaning the end-to-end mapping from low/high-resolution images. On one hand, convolution operation is natural for image processing; on the other hand, non-linear mapping is helpful to explore the non-linear mapping relationship between low resolution and high resolution images and enhance the sparsity of feature representation. The experiments have demonstrated that the proposed convolution sparse network has the ability to restore the detail information from low resolution MR images and achieve better performance for super resolution reconstruction.