A Super-Resolution Image Reconstruction Based On Kalman Filtering And Micro-Saccades

The move of camera platform or the platform vibration generates the movement between the low-resolution image sequence, i.e. uncontrolled micro-scanning. Super-resolution method just uses this aliasing effect to enhance the low-resolution image. But the super-resolution problem is an ill-posed inverse problem. The solution is often trapped in local extremum, thus leading to unlimited solution space and increasing the time complexity. The result is often less than ideal. So we need to use some form of regularization to constrain the solution space or stabilize the problem domain space. But many super-resolution methods with constraints are slow, resulting in poor real-time performance. The poor real-time performance makes many areas’super-resolution problems not be effectively solved, such as remote monitoring, biomedical and small moving target detection and tracking fields etc. Consequently, a low time complexity fast super-resolution has a very important research significance both in theoretical and practical aspects.Given this, we address a novel method, fast feed-back, for super-resolution image reconstruction. This method is based on DKF(Discrete Kalman Filter) and Micro-Saccades. On the one hand, through the analysis of the linear stochastic differential equation of the discrete Kalman filter, we derive our fast feed-back model in this paper. On the other hand, through the analysis of the Micro-Saccade, we interpret our method from the perspective of biomedical in this thesis.Detailed speaking, the proposed approach is similar to the Iterated Backprojection method. But we borrow the concept of Micro-Saccade as a new theoretical source from the human visual angle and deduce our model using DKF from the general computer vision angle. Our method is simple and also functional compared with many other methods on SR inverse problem. It is one realization of bionics.