Research And Realization Of Super-resolution Image Reconstruction Algorithms

Super-resolution image reconstruction is the use of signal processing and imageprocessing, and other imaging methods to eliminate all factors that leading to thewithdrawal of image, and at the same time resume the optical diffraction limit of theresolution decided outside the cutoff frequency information, to form a clear image of thehigher resolution. Generally, the super-resolution image reconstruction method, consistsof four phases: the degradation model, Motion Compensation (including motionestimation and image registration), interpolation and the elimination of blurriness andnoise. Now, a lot of super-resolution algorithms are provided, but these algorithms havetheir own limitations. There is still a lot of research work to do in the super-resolutionreconstruction algorithm.This paper analyzes the basic principles, mathematical models and image qualityevaluation criteria of the super-resolution reconstruction, and summarizes thesuper-resolution image reconstruction methods. Some methods are showed byexperiments. The performance of the various algorithms is compared in this paper.In order to solve the problem of the rebuild algorithm of super resolution images,which have high computing complexity, slow convergent velocity and unstableconvergence, this article presents a robust and regularizing rebuilding algorithm of superresolution. This algorithm creates initial images via wavelet, and improves the quality ofrebuilt images. Measuring regularization using L1 norm, makes the algorithm robust. Itimproves the efficiency of the algorithm with importing adaptive regularizationparameter. This algorithm is effective proved by emulational experimentation.About the noise model choice, in order to simulate the actual noise better, generalizedlikelihood ratio test (GLRT) is to be used to determine the noise model, which can filterthe noise more accurate. On this basis, a non-iterative weighted filtering algorithm(NIWF) is provided in this paper, The algorithm can be used to the robust regularizationalgorithms. There is a good filtering effect to the outlier noise like Salt and Pepper.Set up a simulation platform to simulate expediently super-resolution reconstructionalgorithm, and to improve the efficiency of research and development. The platform isbuilt to a modular design, with good reusability.