Super-resolution Reconstruction Of Video Image

With the development of society,the requirements of video image quality is more and more.Due to the high cost of enhancing the spatial resolution of the image by hardware,enhancing the spatial resolution of image by software,called super-resolution reconstruction technology came into being.In this thesis,we analyze the shortcomings of commonly used super-resolution reconstruction algorithm according to their different theoretical models.First of all,the first step of super resolution reconstruction of video image is image registration.It needs to have the sub-pixel motion estimation,but the robustness of existing algorithms are insufficient;secondly,the sampling process and the defuzzification process is separated in the existing algorithms without using the advantages of image sequences in the defuzzification process.And due to the limitation of the existing algorithms,they can only deal with the integer magnification;finally,the color image processing in the existing algorithms is too simple,so the super-resolution reconstruction results often have the artifacts which will interfere the user's visual observation.This thesis proposes an improved solution to overcome the shortcomings of the super-resolution reconstruction algorithm.The main research work is as follows:(1)Firstly,to overcome the disadvantage of the robustness of image registration in super-resolution reconstruction algorithm,an improved mutual information based image registration method is proposed.In the existing image registration algorithms based on mutual information,they often use the Particle Swarm Optimization(PSO)algorithm or Powell algorithm to optimize the result.But the two algorithms have their own shortcomings.So we combine the PSO algorithm and Powell algorithm.Use the global search ability of PSO algorithm and the local search ability of Powell algorithm to improve the robustness of image registration.And the efficiency of our algorithm is not low.(2)Secondly,in order to overcome the limitation of the super-resolution reconstruction algorithm and the limitation of the amplification,we propose an alternating minimization algorithm.By adding a regularization term which constrains the fuzzy kernel,the algorithm combines the upper sampling process with the deblurring process to make the algorithm more efficient.In order to overcome the shortcomings of the magnification,the original low resolution image sequence is sampled again,so the rational magnification is achieved.(3)Finally,to overcome the shortcomings of color image processing algorithm for super resolution reconstruction,we proposed adding a regularization term which contains the color information to the alternating minimization algorithm for super-resolution reconstruction.The algorithm realizes the effective processing of color image without the artifacts which existing algorithms often appear.Direct observation is more suitable for the user.In this thesis,in order to demonstrate the effectiveness and robustness of the alternating minimization algorithm for super-resolution reconstruction in the practical,we applied of a series of comparative experiments.The results show that the improved algorithm is better than the commonly used algorithms.Both of the subjective evaluation and objective evaluation are higher than other commonly used algorithms.