A camera works to obtain scene information through optical systems and image sensors.The optical system has a cut-off frequency.The image sensor is a downsampling system,which is affected by the density and size of the pixels.Motion blurring during photography and system noise can also cause degradation.Therefore,the camera’s imaging results are degrading digital images.There is a need for high-resolution images in many fields,so people use the super-resolution technologies of hardware or software to improve the resolution of images.Software methods use computer algorithms to improve the spatial resolution of images.Although they have recovery limits,they are low-cost,robust,and flexible.After decades of development,super-resolution technology has many practical algorithms.The multi-frame Spatial algorithms reconstruct continuous low-resolution images sequence with subpixel information.Its representative algorithms are Projection Onto Convex Sets(POCS)iteration,Maximum a Posteriori Probability(MAP),et al.In order to obtain high-resolution images from low-resolution image sequences contaminated by noise,this dissertation focused on the multi-frame POCS/MAP algorithm.The content of this study is to propose an innovative optimization scheme that is conducive to improving reconstruction stability and quality,and reducing the effects of noise while restoring details,based on the existing POCS algorithm,MAP algorithm,and POCS/MAP algorithm.The contents of this dissertation are as follows:(1)This dissertation improved the POCS algorithm by using the bilateral PSF fuzzy kernel in iteration to eliminate the artifacts of the traditional POCS algorithm.For adaptive parameters,this dissertation used the Principal Component Analysis(PCA)noise estimation method based on weak texture area detection and edge strength estimation to estimate the standard deviation of spatial function and the standard deviation of gray value Gaussian function,respectively.To further reduce the impact of noise on the POCS algorithm,this dissertation also extracted high-frequency images with directional coefficient matrix using Non-downsampling Contourlet Transformation(NSCT).It used these high-frequency images for POCS reconstruction.Experiments showed that the improvements in this dissertation could remove artifacts and improve the signal-to-noise ratio of the reconstructed POCS.(2)Based on the study of Bayesian theory and prior models,we proposed a new combination of prior models.This prior model consists of two parts: the first is the gradient term on low-scale images’ up-sampling,introducing cross-scale prior information and having a smooth effect.The second is the piecewise differential term used to preserve local details.This dissertation introduced a luxury variable to solve the non-convex problem of this prior combinatorial model.It used the L2 norm and L1 norm alternating optimization method to solve the stable convergence results.(3)This dissertation proposed a new combination of prior models for the POCS/MAP hybrid algorithm based on separating the components separated by the mixed sparse representation(MSR)framework.This dissertation used the MSR to separate image components and model the smoothing and edge components separately.The smoothing component is a Gauss-Gibbs model for noise suppression,and a Lorentz model approximates the edge component.Experiments showed that the proposed algorithm is effective and robust under different noise conditions.This study is suitable for the obtained image is polluted by noise.When detecting unknown targets in harsh environments,the image may be polluted,and it is not easy to get many sample images.Using the algorithm studied in this dissertation can suppress noise while obtaining clear images.In the case of limited camera hardware design,the algorithm proposed in this dissertation can recover some information beyond the cut-off frequency and effectively improve the clarity of the image. |