Research On Super Resolution Image Reconstruction Technology Based On Compressed Sensing

High-resolution images can fully represent target features,provide richer scene information and are good for the analysis,understanding and recognition of the targets.However,during the actual imaging process,optical distortion,motion blur,undersampling,random noise and other factors eventually reduce the spatial resolution of the imaging system.Compressed sensing(CS)as a new sampling theory,can perform compressive sampling on sparse signals or sparse signals in a certain transform domain,and then uses the compressed sampling signals to reconstruct original signal with high probability and accuracy.The theory of recovering the original signal(high-dimensional signal)from a small number of sampling points(low-dimensional signal)undoubtedly provides a new idea and method for image super?resolution reconstruction.Based on this,this paper will focus on the research of image super?resolution reconstruction based on compressed sensing theory.The main contents are as follows:Compared with visible image sensors,the spatial resolution of infrared image sensors is low.Thus,in order to improve the spatial resolution of infrared images,we propose a single?frame infrared image super-resolution reconstruction algorithm based on over-complete d:ictionary sparse representation.According to the reconstruction model,the proposed algorithm can be divided into two different types,one based on image statistics and CS,and the other one based on image degradation model and CS.The method based on image statistics and CS needs to obtain a pair of over-complete dictionaries to describe the relationship between high resolution and low resolution image blocks.The method based on image degradation model and CS only needs to train to get one single dictionary.In order to make the algorithm have higher reconstruction efficiency and accuracy,the algorithm uses a nonlinear filter to perform basic and detail separation operation on the image.Experimental results show that the proposed algorithm effectively improves the spatial resolution of infrared images,and the PSNR index of the high-resolution infrared image reconstructed by the method based on image degradation model and CS is 1.992dB higher on average than the method based on neighbor embedding.Because single dictionary(or dictionary pair)cannot describle complex mappings,a single-frame infrared image super-resolution reconstruction algorithm based on clustering multi-dictionary mapping is proposed.The algorithm firstly uses the clustering method based on multi-scale principal component analysis to classify the image block samples according to their main geometric directions,and then multiple sub-dictionaries can be obtained by training these datasets.Compared with a single dictionary mapping,multi-dictionary mapping ean transform a complex nonlinear mappinging into multiple simple linear mappings,which reducing the artifacts in the reconstructed images and improving the reconstruction performance.In addition,the proposed algorithm introduces the image degradation model and the improved non-local mean filter as two regular constraints for the secondary correction for the preliminary reconstructed high-resolution image,which not only guarantees the ability of the algorithm to suppress random noise,but also further improves the performance of the algorithm.The experimental results in our article show that the PSNR index of proposed method is 0.628dB higher on average than the method based on image statistics and CS.For single pixel camera.due to the performance of improving the spatial resolution is limited,thus,a super-resolution ilaging lethod based on dynamic compressed sensing theory is proposed.Compared with traditional single-pixel imaging systems,the proposed method can obtain higher spatial resolution images and has an obvious advantage in reconstruction time.In order to obtain images with sub-pixel displacements?we load patterns with sub-pixel displacements on DMD device,which makes the proposed method to achieve sub-pixel displacement without the need for additional hardware devices.Besides,we propose a lulti?sampling filtering technique that can effectively improve the signal-to-noise ratio of the measured values.Simulation and experimental results demonstrate the feasibility and effectiveness of the proposed method.For traditional single pixel imaging system,the efficiency of getting measured values is very low.Based on this,a super-resolution imaging method based on parallel compressed sensing theory is proposed.The method introduces random scattering media into the imaging system and makes it as the measurement matrix of the system,so that all the measurements can be obtained in a single measurement.which achieving parallelization of the data acquisition and greatly improves the acquisition efficiency of measured values.The proposed systen can implement imaging for moving targets.which overcomes the defect that traditional single-pixel imaging system cannot realize it.In the reconstruction phase,we propose a two-step phase shifting technique.This technique can transform the complex phase recovery problem into a standard compressed sensing reconstruction problem.In combination with the TVAL3 algorithm,the final high-resolution image can be reconstructed.Simulation and experimental results demonstrate the feasibility and effectiveness of the proposed method.In this paper.the image super-resolution reconstruction algorithm based on compressive sensing theory is studied,and new reconstruction algorithms and new imaging mechanisms are proposed.Above researches have achieved initial results.Above studies can provide good technical support for low-resolution imaging system to obtain high-resolution images,and provide a theoretical basis for new imaging systems.