Optimization Approach to Super-Resolution Image Reconstruction

Introduction

The process of super-resolution is an inverse problem of estimating a high-resolution image from a sequence of observed, low-resolution images and it is now widely known to be intrinsically unstable or “ill-conditioned”. The common feature of such ill-conditioned problems is that the small variations in the observed images can cause (arbitrary) large changes in the reconstruction. This sensitivity of the reconstruction process on the input data errors may lead to the restoration errors that are practically unbounded. The important part of super-resolution process is thus to modify the original problem in such a way that the solution is meaningful and a close approximation of the true scene but, at the same time, it is less sensitive to errors in the observed images. The procedure of achieving this goal and to stabilize the reconstruction process is known as Regularization.

Many super-resolution techniques [8, 10, 16] are based on the optimization approach. We re-examine the approach to super-resolution using optimization techniques by reformulating the problem in terms of a regularized optimization procedure and implement an iterative conjugate gradient method for finding the minimum of the resulting objective function. The role of regularization term on the accuracy of the super-resolution reconstruction is also investigated. For evaluating the performance of the IISR technique, the full-solution of IISR is compared with the one generated by the optimization technique [46].