Research On Blind Restoration And Super-resolution Reconstruction Algorithm For Image Clarify

In many application, images are degraded in some degree because of various reasons.However, people need to get higher quality images in practical situations.So,how to effectively recover the original image, or recovery out the image that we expected, become an important question in image processing. In this thesis, the research work is based on image blind restoration and super-resolution reconstruction algorithm, which aims at researching image blind restoration methods to estimate original undistorted image by means of known or unknown partial prior knowledge about blurred image itself and PSF, also, researching super-resolution reconstruction, trying to break through the restriction of hardware device and get a high resolution image.This dissertation will focus the image blind restoration and super-resolution reconstruction algorithm carried out research, which includes the following parts:image degradation models; parameters estimation of motion blurred images based on cepstrum method and blind restoration of motion image; total variation minimizing iterativeness blind image restoration; maximum a posteriori algorithms.The main work and innovations in the thesis are listed as follows.1. Discussion is given on the development history and study actuality of the image blind restoration and super-resolution reconstruction technique and explanation is illustrated on the significance of the thesis.2. In view of motion blurred images, the thesis proposes cepstrum method to estimate PSF. The method incorporated the cepstrum properties of motion blurred images with the mathematical significance of Radon transform, efficiently estimating the motion-blurred parameters.3. In the algorithm of total variation minimizing frequency iterativeness blind image restoration, in order to reduce the complexity of this algorithm, the thesis simplifies the choice of regularization parameters. The algorithm can be easily implemented.4. In the dissertation, the thesis studies the super-resolution reconstruction algorithm. Based upon the detailed analysis on the optical flow estimation approach properties, an optical flow method with the motion oriented smoothness constraint is proposed, which can enhance the accurateness of motion estimation results in the case of small geometric distortions. Then, solution space was constrained by exploring Markov Random Field, based on the MAP frame model. Meanwhile, through using Gibbs Random Field, so redundant information between image pixels was made explicit use of to construct high-resolution images, and improving the resolution.