MAP-based Technology Study Of Image Super-resolution Restoration

The goal of Super-resolution restoration is to draw a high-resolution image from a low-resolution ratio image array. Super-resolution refer to removal the image system (out of focus blur, motion blur, non-ideal sampling, etc.) and recovery of clear image that got after resuming high-frequency information of the formation of image with lost optical system. Although classical linear image restoration (such as linear and cubic spline expansion) has been thoroughly studied and implemented, because the edge area of the image and discontinuous area become fuzzy.So study in guarantee image Super-resolution ratio result and is it test prerequisite that speed does not reduce to charge, how establish image model, reduce computer operation amount, is it have important meaning and value to recover as to image. This text realizes from design and algorithm of the model to this question that two respects have carried on deep research and realized. The groundwork and innovation of the thesis include:First, Super-resolution re ratio deficiency that exist in the method to single width image, design one test probability estimate(MAP) under the frame because of being most heavy, utilize the high-resolution image of the array image of many frames to construct the method again. MAP estimates that has offered the good frame for the optimum estimation of the image, convenient mathematics is asked and solved, can obtain it than the image information with more single width image, construct out the resolution ratio with higher more image than any single frame again.Second, is it choose one Gibbs model to recover to image, design his parameter choose mode. Through Gibbs model, can analyze the whole composition of the image from some characteristic of the image, having good convergence efficiency, this has brought convenience for real-time processing.Third, considering the characters of common moving image, this thesis puts forward to an improved three step searching method, which can deal with the image obtained from in any situation either with comparatively bigger movement or with comparatively smaller movement. This method has a good effect, when it used in the smaller movement estimation. Also it speeds up the process of the movement estimation and diminishes the calculation of Super-resolution restoration by combining such a feature that the most superior point in the moving image usually distributes around the zero vector.Forth, we use an improved algorithm based on linear spline expansion in the first step of recursive method, which deals with the deficiency of the linear spline expansion when it come to the verge process. Through the improvement, the detail of the expansive image is still clear.This thesis reconstructs high resolution image with better quality compared to the existing methods and better efficiency than other similar methods. Finally, it is proved by the experiments.