Research On The Algorithm Of Super-resolution Reconstruction

Multiframe super-resolution reconstruction is an active topic in image processing which is urgently needed in many military and civilian applications as well as theoretical interests. This dissertation focuses on the super-resolution from a sequence of low-resolution. The study is carried out in the following aspects:We study multiframe super-resolution reconstruction for ordinary model. The algorithm of Tikhonov regularization, TSVD, Landweber iteration and CG are introduced to solve ill-posed inverse problems, among which CG is emphasized because of its quick convergence.Boundary condition influences computational complexity and boundary errors. Compared with zero and periodic boundary conditions, Neumann boundary is more satisfactory. Generally speaking, image restoration is globally treated, which may lead to a huge or infeasible problem. So local deblurring is introduced. It divides a big noisy and blurred image into small blocks and restores them individually to produce the restored image. Experimental results show the efficiency of appropriate blocking in reducing computational cost and achieving a close performance to that of global algorithms.The problem of image interpolation is investigated. Due to the existence of quantitative errors, interpolation techniques should be utilized to avoid overlaps and holes when constructing the warp matrix in super-resolution. Compared with linear and neighbor interpolators, cubic interpolation can attain much better reconstruction results. Regarding the inherent drawbacks of linear interpolation techniques, non-linear interpolation is also introduced.