Super-Resolution Technique for MRI Images Using Artificial Neural Network

Abstract

Image upscaling is an important field of digital image processing. It is often required to create higher resolution images from the lower resolution images at hand in computer graphics, media devices, satellite imagery etc. Upscaling is also referred to as ?single image super-resolution'. The process is a tradeoff between efficiency, time and the quality of output images obtained. Images with higher quality are needed and are essential in many areas like medical, astronomy, surveillance, satellite imaging etc. In medical imaging, images are obtained for medical investigative purposes and for providing information about the important diagnosis instrument to determine the presence of certain diseases. Many techniques like PET (Positron Emission Tomography), CT (Computed Tomography), MRI (Magnetic Resonance Imaging) in the medical field are used for detecting diseases. Generally, medical images suffer from low resolution, High level of noise and blur type of factors. In the present paper, a feed forward neural network using supervised training for image upscaling is proposed. The performance of a neural network is compared to different training function & measure PSNR.