Factor transfer for MRI image super-resolution

In order to obtain clear MRI images with high resolution, super-resolution is an important technique in the current image processing field. The current deep learning in the field of image super-resolution has been proved to be effective, but the recovery of MRI images with noise is still unsatisfactory, often with artifacts and blurring. In this study, we propose a MRI image super-resolution model called FTSR that maximizes the transfer of the teacher's "factor". The FTSR model maximizes the knowledge obtained from the teacher's model through knowledge distillation and transfers it to the student's network, so as to guide the students to complete the super-resolution reconstruction of under-sampled MRI images with noise.