Research On Image Super-resolution Restoration

Recently, more and more image processing experts pay attention to imagesuper-resolution technology,thanks to its possible perfect future of application.Generally speaking, image super-resolution restoration (reconstruction) and imageresolution enhancement. could belongs to image super-resolution technology.conceptually, the former is based on image restoration and reconstruction ,and thelatter covers the former, even covers image fusion of sequence frames of single sensor. The terminology "resolution"represents the number of pixels in an image, whichdetermines the physical size of the image, and also the fidelity to the high frequencydetails in the image. Conventional interpolation algorithms, such as zero-order ornearest neighbor, bilinear, cubic B-spline interpolation, focus on just enlargement ofimage. Those algorithms have been developed under assumption that there are nomixture among adjacent pixel in the imaging sensor, no blur ,distortion and noise in theprocess of getting low-resolution image from high-resolution image. Image restorationis focus on the reason of all types of distortions in the imaging system, and highlyrestores the truth image. The problem of image super-resolution restoration, which need integrate imageinterpolation and image restoration, is more ill-conditioned and ill-posed than that ofimage restoraion. However the technique of image super-resolution restoration makesit possible that high resolution images could be restored from low resolution imagesrecorded by low resolution sensors. super-resolution restoration algorithms may bedivided into two classes, particularly frequency domain and spatial domain. Allfrequency domain approaches made use of the aliasing effect; while spatial domainalgorithms have mainly three approaches, i.e. Iterative Backward Projecting(IBP),Projection Onto Convex Sets(POCS) and Bayesian methods. In this paper, a parallelgenetic framework algorithm for image (sequence) super-resolution restoration ispresented . The parallelism of this real-valued genetic algorithm based on the islandmodel enables better integration of the information of the multiple frame images.Especially with the iterative method of other super-resolution algorithms being themutation operator, the convergence of the genetic searching in the solution space is fast.The experiments demonstrate that the proposed algorithm is efficient and applicable. Further more, considering the identification of blur of imaging system, weproposed a novel regularized image interpolation algorithm based on GeneticAlgorithm. This genetic algorithm also has real-valued coding, the induced mutationoperator and the fitness function for evaluation containing the term of some subjectivequality measures, so the convergence of the genetic searching in the solution space isvery fast. Finally, then we analyzed how to choose the regularization parameter in thefitness function, and compared the results with that of iterative regularizedinterpolation algorithm. The experiments demonstrate that the proposed algorithm ispractical and applicable.