Research For Super-resolution Image Reconstruction Algorithm Based On Regularization

Super-resolution image reconstruction (SRR) refers to a resolution enhancement technology that extract higher resolution images containing more details from an image sequence of lower resolution by using digital signal processing technology, and the existing low-resolution imaging systems can be still utilized. SRR has a wide range of applications and has become one of the hottest image restoration research topics in the world. The fast and efficient reconstruction algorithms, and the high-precision movement registration algorithm are the focus of the study and difficulties of Super-resolution reconstruction.In this paper, based on the image reconstruction model and space domain approach, the MAP SR image reconstruction method is mainly researched. The basic principal of this algorithm is introduced, and the selection of the cost-function and regularization term has been deeply analyzed and compared by experiment. Considering the reconstruction result and iteration speed will be affected by the LR image we used,A new cost function is presented under the L1-norm reconstruction framework, adds the data fitting term and regularization term of the missed low-resolution images to the cost function, and solved it by alternating minimization method. Experimental results demonstrate the effectiveness of the proposed method in fewer low-resolution images observed condition.The regularization parameter plays an important role in reconstruction process. Traditionally,the choice of regularization parameter always based on experience, by comparing the different results of experiment. In this paper, detailed analysis and discussion of the selection of the regularization parameter has been given. And based on the existing methods, such as L-curve and proportion method, we proposed a dynamic update selection method of the regularization parameter by using the segmented function. In this method, according to the changes of the fitting term and regularization term between two adjacent iterations, we choose the different function to calculate the parameter. In this way, it will not only guarantee the rational calculation of the parameter, but also can improve the speed of the convergence.