Magnetic Resonance Imaging (MRI) can provide anatomical images of internal organs to facilitate early diagnosis of the disease. But the inherent defects of medical imaging system make the acquisition of HR medical images face many problems. One way to solve these problems is to use super-resolution reconstruction technique. We design a feed-forward full connection convolution neural network, which includes five convolution layers and five residual blocks. In addition, loss function based on perception is also used to solve the problem caused by mean square error loss function which cannot meet the human visual sense very well. This method realizes build-in 4 times magnification reconstruction and avoids the checkerboard artifacts, which are often occurred when using deconvolution layers to up-sample images in convolution neural networks (CNN). The effectiveness of the method is verified by experiments, and both the visual and numerical results are improved.