Research On High Spatial Resolution Of Planar Compound-eye

Planar compound-eye imaging system has the advantages of compact structure and diverse functions,and has a wide application prospect in the fields of imaging equipment miniaturization and computational imaging.However,due to the small sub-aperture of the compound-eye and the low spatial sampling rate of the image sensor of the sub-aperture,the image quality and spatial resolution of the sub-aperture is low.How to improve the spatial resolution of the planar compound-eye is an urgent problem for high resolution application scenarios.Although a variety of high resolution image reconstruction or multi-image super-resolution algorithms have been used to improve the spatial resolution in existing planar compound-eye systems,high resolution reconstruction algorithms adopted by these systems still have the problems of poor adaptability,robustness and small application range,which greatly limits the spatial resolution,the flexible design and practical application of the high resolution planar compound-eye imaging system.In order to solve the above problems,this dissertation focuses on the improvement of the spatial resolution of the planar compound-eye system,and uses variational Bayesian multi-image super-resolution as the theory framework to improve the spatial resolution of the planar compound-eye system.We focus on how to improve the adaptability,robustness and resolution enhancement performance of the multi-image super-resolution algorithm and corresponding planar compound-eye system.The main research content of this dissertation can be divided into the following three parts:1.Due to the importance of the high-resolution image prior model for multi-image super-resolution for improving the spatial resolution of the planar compound-eye,an image prior model based on filter bank and l₁ norm is proposed.The filter band of the prior model can be selected and designed flexibly and the sparse l₁norm has better high-frequency detail restoration performance.Then,a multi-image super-resolution algorithm based on the proposed prior model is obtained under the framework of variational Bayesian framework.Planar compound-eye system based on this algorithm has better adaptability and high spatial resolution performance.The performance advantages of the proposed high-resolution image prior model and the corresponding multi-image super-resolution algorithm are further verified by the compound-eye simulation data and the prototype data experiments.2.For the first time,the Laplace distribution noise observation model is applied to the variational Bayesian multi-image super-resolution theory,and the parameters in the observation model are modeled under the hierarchical Bayesian framework.Corresponding multi-image super-resolution algorithm is deduced under the variational Bayesian multi-image super-resolution framework.The planar compound-eye system using this algorithm to improve resolution has the characteristic of local noise adaptive,better robustness to impulse noise and outliers,and better parameter adaptive.The spatial resolution performance and robustness of the proposed method are fully verified by a wide range of compound-eye simulation data experiments and prototype data experiments.3.The existing multi-image super-resolution theories adopted in the planar compound-eye system usually have the following assumptions:1)the sub-aperture blur kernel?i.e.point spread function?is the same and known,and 2)the relative motion model between sub-aperture images is modeled by Euclidean transformation.These two hypotheses greatly limit the application range,flexible design and active control of the planar compound-eye imaging system.In order to solve this problem,in the variational Bayesian multi-image super-resolution framework,we incorporate the blur kernel statistical modeling and estimation,and apply the affine transformation,which take Euclidean transformation as a special case.At the same time,combining the high-resolution image prior model and the robust observation model proposed in this dissertation,a robust,parameter-adaptive and more widely applicable high-resolution reconstruction algorithm for the planar compound-eye is finally obtained.This algorithm not only improves the spatial resolution of the whole planar compound-eye system,but also provides important theoretical support for the more flexible system design,active control and broader scope of application of the high-resolution planar compound-eye system.A large number of compound-eye simulation data experiments and prototype data experiments have verified the correctness and effectiveness of the proposed relative motion model,blur kernel estimation and the whole proposed high-resolution reconstruction technology.