Super-Resolution Image Reconstruction with Adaptive Regularization Parameter

The super-resolution reconstruction can be regarded as a typical ill-posed inverse problem. Regularization method is the most important method used to solve this kind of problem. How to determine the regularization parameter is the most critical and most difficult problem in the regularization algorithm. We propose a method for adaptive determination of the regularization parameters for super-resolution Image reconstruction. The proposal relies on the structure tensor. Besides using traditional mathematical methods of ill-posed inverse problems, this method pays more attention to the image structural characteristics of smooth, angular, edge and others. We determine regularization parameter adaptively that the parameter values is small at the edge and texture and other non-smooth regions, especially angular, and in the smooth, uniform blocks, the pixels corresponding to the parameter value is large. We contrast the proposed method to the classical methods such as Tikhonov regularization, GCV, L-curve. Experimental results are provided to illustrate the effectiveness which makes regular of the role of the reconstructed image intensity changes in the degree of local smooth adaptive to change, help to protect the image detail, while smooth regions to better noise suppression.