Blind Super-resolution Image Reconstruct Based On Ggenetic Algorithm And Regularization

In the process of capturing image, the original scene from the camera is easily warped because of the scene motion and the camera position change. In addition, it is difficult to obtain the clear high-resolution (HR) image due to various reasons, such as atmospheric turbulence, camera lens and the connatural limitation of digitized image sensor, etc. Base on the existing hardware conditions, how to acquire the HR from the low-resolution (LR) image would be the core of the Super Resolution (SR).The SR technology was developed nearly 50 years ago. The application of the early SR approach based on the signal frame image is limited because there is little helpful information could be used. The SR approach based on the multi-frame was proposed in 1980's. It contains two approaches: frequency approach and space approach. As the classical SR model was founded, more and more attention has been devoted to study the space approach. In generally, it contains three elements: Registration, Interpolation and Remove the blur. Registration is the precondition of the SR technology. The accuracy of Registration is directly relation to the quality of the restoration image. The image interpolation and image restoration are integrated, which means to recover image on super-resolution grids. And there are more ill-conditioned and ill-posed than that of image restoration. The frame of regularization is introduced to restrict the ill-posed problem At the same, the genetic algorithm which is of better searching ability of global optimal solution is introduced to carry out regularization interpolation between the frames of image sequence.However, traditional technology of SR reconstruction assumes that the spread function of imaging system is a known point, and then handles the LR images for the HR image. How to achieve the blind super-resolution without any priori known blurs will be the new problem we faced. For overcoming these problems, the approach of blind super-resolution image reconstruct based on genetic algorithm is proposed; this approach introduces a regularization energy function, where regularization includes both the image and blurs domains. In order to enhance efficiency of the HR reconstruction and the quality of the image, we minimize the regularization by using global optimizing capacity of genetic algorithm. it achieve the reconstruction of blind super-resolution. In addition to that above, the rational down-sampling factor is introduced into the traditional integer down-sampling factor in this paper. The process of the rational down-sampling factor to be uses in the reconstruction of HR image is presented. The relation between the number of the LR images frames and SR factor is also illuminated. The results of the simulation is analyzed and compared with those of the traditional approaches. The simulation results demonstrate that the proposed algorithm is practical and applicable.