IEICE Information and Communication Technology Forum
Patch based digital image processing - Principles and selected applications
PDF download (278.7KB)
We are witnessing rapid emergence of patch based, non-local digital image processing techniques. During the last decade, these techniques led to tremendous improvements in denoising, deblurring, inpainting, superresolution and many other image processing tasks. By modeling and manipulating patches of image pixels and by considering and grouping similar contexts non-locally, throughout the image, these approaches improve strongly over the more traditional pixel-based and local techniques. In this talk, we review some of the basic principles of patch-based image processing, from modeling and grouping of self-similar patches to learning dictionaries of patch-based image atoms. Selected applications in image restoration, super-resolution and inpainting will be discussed.