A New Method for Super Resolution Reconstruction'

In this presentation, a new algorithm is proposed for super resolution construction by combining NLM and BM3D methods with Tikhonov regularization. First a summary of principles and a brief review of the two methods (NLM and BM3D) on noise reduction are presented and their excellent performances on noise suppression are showed. Second, through an insight understanding of the least square solution of the general ill-posed problem (AX=B, where A has small eigenvalues), a new Tikhonov regularization combined NLM is introduced to reconstruct a super resolution image for single frame case. The regularization adjusts the balance between structure restoration and noise suppression with a weighted penalty. At last, we use BM3D to reduce noise that was amplified in the result of Tikhonov regularization. This new methods can lead to good results whose PSNR is about 29 in the case of small additive noise. There are also some advantages introduced in the conclusion part.