Research On Image Super-Resolution Reconstruction Algorithm Based On Regularization

Image super-resolution reconstruction is an important technology in image reconstruction field. In the process of collecting and processing image, various factors, including hardware equipment and outside objective reasons, could affect the image system. Therefore, the quanlity of reconstructed image will be declined. And, it is becoming more and more difficult to increase image resolution by upgrading the traditional hardware methods. Therefore, it is becoming more and more important to increase image resolution by software processing method. Facing the situation, researchers adopt image super-resolution reconstruction method to solve the problem. The concept of image super-resolution reconstruction is that utilizing a series of low-resolution images to reconstruct a high-resolution image by redundant and complementary information of low-resolution images on the same scene.The progress of image super-resolution reconstruction is a typical ill-posed problem. The regularization method is a good method to solve its ill-posed problem. And the regularization method is the mainstream method for image super-resolution reconstruction. Therefore, the emphasis of this thesis is the image super-resolution reconstruction based on the regularization method.Firstly, starting from linear ill-posed problem, some typical regularization methods are researched, then the effect of the choice of regularization parameter on reconstruction process is analyzed, and some common choice methods of regularization parameter are researched. The traditional regularization could not update iteration step adaptively in the iteration process, and the relevant information of different image channel could not be utilized. Facing the situation, this thesis proposes a new method that combines updating regularization parameter adaptively and updating iteration step adaptively. Finally three group simulation experiments of three images are carried on to demonstrate the proposed method. The experiment results indicate the proposed algorithm can retain the original image edge information and improve the quality of reconstucted image.At present, the research emphasis of regularization method is the choice of regularization term. But, most of regularization term is based on spatial field. In other words, the researchers only assume that image scence is static and spatial resolution is limited. The time resolution of dynamic image scence is not taken into account. That is, the prior constraint information in time domain is not considered. Facing the issue, a new regularization method is proposed. The new method considers introducing the regularizaiton term of space domain and time domain simultaneously. At the same time, the method defines the adaptive weight of regularization term of space domain and time domain, which enhances the adaptivity of the new proposed algorithm. The simulation experiments indicate that the new regularization image super-resolution reconstruction algorithm based on combining space domain and time domain could improve the stypticity and stability in the iteration process and improve the quality of reconstruction image.