Single Image Super-Resolution via Edge Reconstruction and Image Fusion

For decades, image super-resolution reconstruction is one of the research hotspots in the field of image processing. This paper presents a novel approach to deal with single image super-resolution. It's proven that image patches can be represented as a sparse linear combination of elements from a well-chosen over-complete dictionary. Using a dictionary of image patches learned by K-SVD algorithm, we exploit the similarity of sparse representations to form an image with edge-preserving information as guidance. After optimizing the guide image, the joint bilateral filter is applied to transfer the edge and contour information to gain smooth edge details. Merged with texture-preserving images, experiments show that the reconstructed images have higher visual quality compared to other similar SR methods.