Super-Resolution Based On Over-Complete Dictionary

Image Super Resolution (SR) aims to recover a high-resolution (HR) image from one or more low-resolution (LR) images. In recent years, it is one of the most active research topics in digital image processing, face recognition, military and so on. SR techniques are very important both in theory and real applications. Currently, numerous super resolution algorithms mainly focus on learning the prior knowledge by a large of sample images. The kind of algorithm not only heavily depends on the sample images, but also the computing speed is slow. Furthermore, if magnification becomes larger, they are more prone to produce ringing and jagged artifacts.The main research of this paper focuses on the image super-resolution by learning the over-complete dictionary. This paper aims to reconstruct the high-resolution image effective and reliable by the proposed algorithms. The main contents are as following:(1) This paper introduces the background, numerous current SR methods, the dictionary learning and sparse theory. Furthermore, it also analyzes the status and the development of SR algorithms.(2) Based on K-SVD approach, the paper proposes the R-KSVD algorithm, which obtains dictionary effective at fast computing speed. This paper applies R-KSVD algorithm to learn the dictionary without any extra dataset, and then recover a high-resolution image by the dictionary. The proposed method obtains the effective HR image in visual and RMSE.(3) Taking the structure of the LR image consideration into SR algorithm, the paper proposes a novel SR algorithm. It called Laplacian sparse coding for super resolution. (4) This paper applies bilateral filter to preserve the edge information and reduce the reconstruction error, then sharp edge can be achieved.