Research On Spatial Image Sequence Super-resolution Reconstruction And Its Robustness

With the development of multimedia communications and information processing technologies,people are increasingly demanding image resolution.However,many imaging systems are limited by physical conditions,and the resulting images often have low resolution,and the visual effects of the images are difficult to meet people's needs.If the method of improving physical hardware is used to obtain a high-resolution image,the cost is high,and it is sometimes difficult to achieve.Therefore,it is necessary to consider a new method that can overcome these limitations to improve the spatial resolution of the image.Super-resolution reconstruction technology came into being in this context.This technology makes up for the lack of hardware,enhances the resolution of the image from the software's point of view,and enhances the usability of the image.Therefore,as soon as this technology is proposed,it has attracted the attention and attention of many scholars.At present,it has important practical application value in television,remote sensing,medicine,and public security.Super-resolution image reconstruction refers to the process of processing multiple low-resolution images with global or local sub-pixel displacements between each other to obtain higher resolution and better imaging images or videos.Based on a single frame,multiple frames of still scenes and sequences of three major categories of frames.Among them,the super-resolution reconstruction technology based on sequential multi-frames is a difficult and difficult point in the current research and needs to be further studied.This thesis discusses the development and basic theory of the super-resolution reconstruction technique of sequence images,and systematically studies the various steps of super-resolution reconstruction of sequence images.Focusing on the inter-frame registration of sequence images,spatial domain reconstruction methods,and super-resolution reconstruction techniques based on canonical models,this thesis carried out theoretical simulation,algorithm improvement,and experimental verification,and achieved certain research.Achievements.The research content and research results of the dissertation mainly include the following:1.Based on keren registration method,phase correlation method and phase-dependent discrete up-sampling method,simulation and experimental verification are performed.The measured accuracy of registration can reach 0.15-0.25 pixels,which can support subsequent high-resolution reconstruction.2.The full variational regularity and the bilateral filtering regular reconstruction method are compared and compared.The experimental results show that the reconstruction method based on the bilateral filtering regularity performs better in maintaining image edge and robustness,and discusses the difference in the bilateral filtering regular reconstruction method,the influence of the selection of regular parameters on the reconstruction result;3.Experimental verification of the three spatial domain reconstruction algorithms for backward iterative projection method,convex set projection method and super-resolution reconstruction based on bilateral filtering can be performed.The experimental results show that the convex set projection method and the super-resolution method based on bilateral filtering can distinguish Improve and improve image quality have certain advantages;4.The study of super-resolution reconstruction based on deep convolutional neural network can be used to improve the convergence of spatial domain super-resolution image reconstruction.In summary,this thesis analyzes the related basic theories involved in the super-resolution reconstruction of sequence images at the current stage,surrounding inter-frame registration of sequence images,spatial domain reconstruction method,super-resolution reconstruction technology based on canonical model,etc.The key issues were explored and milestones were achieved.The related results of the dissertation provide a theoretical basis for further algorithm research and have certain guiding significance for improving spatial resolution and detail definition of the image.