Image extrapolation using K-SVD algorithm

This article presents a novel algorithm for image extrapolation, which utilizies K-SVD algorithm for training the overcomplete dictionary. With the trained dictionary we can get the extrapolated part of the image. The main idea is to perform the extrapolation for only one row or column at a time, and to drop some constraints on the overlapping of the patches. Our proposed algorithm is compared with the algorithm for image inpainting which can be also used for image extrapolation (with some special conditions). At the end some examples of extrapolated images are shown.