High Quality Image Magnification Using Cross-Scale Self-Similarity

In medical imaging there is a frequent need to magnify a certain region of interest (ROI) of an image. However many modalities suffer from severe noise and traditional upscaling methods produce poor enlargement results. We present an approach based on cross-scale similarity of an image extendable to sequences using time domain information without explicit motion compensation. It combines noise suppression and upscaling and hence yields superior image quality compared to standard zooming algorithms. We evaluated our algorithm using artificial and clinical fluoroscopy acquisitions.