The Research Of Image Super Resolution Reconstruction Algorithm Based On Dictionary Learning

Due to the limitations of technology, volume, price of hardware, researchers are mainly focused on software method to improve the resolution of the images. As an active field in image processing, super-resolution has been widely applied into computer vision, image remote sensing, military image processing, medical imaging and security detection. It is a technology to improve image quality, its purpose is to make the degrade image recover to the real appearance of the original image as much as possible. Image super-resolution aims to reconstruct high resolution image from one or multiple low resolution image(s), high resolution means that the image has the higher pixel density and can provide more details. With the flourishing of digital image, super-resolution has attracted more and more researchers attention due to its extensive applications in real world. Single-image super-resolution is a severely ill-posed problem, the solution to which thus requires effective prior information as a supplement. At present, the mainstream of image super-resolution reconstruction method can be roughly divided into three categories, interpolation based method, reconstruction based method and learning based method. Among which, the learning based method is the hot spot in recent years.On the basis of summarizing the characteristics of three kinds of methods, the paper further study of the principle and framework of the image super-resolution algorithm based on sparse representation, mainly studied the method of the dictionary structure, and optimize the structure of the dictionary on this basis, proposed two new method of the dictionary structure for image super-resolution algorithm.The main research content is as follows:Firstly, this paper introduced the currently used super-resolution reconstruction technologies generally, and detailed the main idea of each method. This paper highlighted the learning based super-resolution reconstruction algorithm, and studied two typical algorithms among these, namely SCSR algorithm, SSSR algorithm.Secondly, the paper constructed dual-mixed dictionary. Among them, the first layer dictionary uses semi-coupled dictionary, which ensures the flexibility and accuracy of the recovery process, and combines with sparse representation algorithm to get the first layer of the restored image. In order not to affect the overall computing speed, the second layer dictionary adopts classification dictionary, and uses the difference between the high resolution natural images and the corresponding low-resolution images as the high resolution sample to restore more high frequency details. The experiment results show that the proposed algorithm effectively improves the quality of the images.Finally, the kernel fuction was introduced into the sparse coding technique, proposed the kernel sparse representation based of image super-resolution reconstruction algorithm. The introduction of kernel function can be more easily to get the image sparse coding. it is not only can reduce the quantization error of features, but also can enhance the performance of sparse coding. The experiment results show that kernel sparse representation algorithm is superior to the tradictional sparse representation algorithm in the field of image super-resolution renconstruction.