Research On High-resolution Image Reconstruction Based On Phase Encoding Diffration Imaging

With the development of modern information technology,people have higher and higher requirements for imaging systems.Not only higher resolution,larger field of view,and longer depth of field are needed,but also lower cost,smaller size and simpler structure.In order to meet the requirements of imaging performance,traditional optical systems often increase the size of the system,introduce aspheric components,or increase the complexity of the system,which results in a large system,a complex structure,and a difficult assembly.In response to this problem,many new methods such as diffractive optical imaging,metasurface imaging,and computational optical imaging have become research hotspots in recent years.They have fundamentally broken through the traditional imaging mechanism to meet the needs of different applications.Phase-coded diffraction imaging devices and methods are a new type of imaging technology that combines the principles of diffractive optics and computational imaging.This technology modulates the incident light through an optical element that can encode and calculate optical information.In combination with post-image restoration technology,high resolution imaging can be achieved using only a single element or a very simple optical module,greatly reducing the optical system.Complexity.Different from the traditional optical system object-image mapping imaging mode,the optical system part of phase-coded diffractive imaging can encode and calculate the incident light information,and consider the entire optical hardware and back-end computer processing software as an information processing module.At this time,the image collected by the sensor is not a clear image that can be directly observed by the human eye,but a fuzzy coded image.It needs to be decoded and restored by computer digital image processing technology to obtain the final high-quality image.The imaging quality of phase-coded diffraction imaging system is closely related to the image restoration algorithm.Since the development of image processing technology,a large number of classic restoration algorithms have been formed,including inverse filtering,Wiener filtering,full variational models,and regular filtering.These algorithms have been applied in the previous phase coding imaging technology research,but most of the current restoration algorithms are the default optical system as a linear space invariant system.All the encoded images are calibrated to the optical system through basic calibration parameters such as the point spread function.The modulation model is characterized and the image is reconstructed.The large-field-of-view and wide-spectrum images obtained by actual phase-coded imaging systems are essentially non-linear space-invariant systems,and the point spread functions from the focal point or off-axis point are not exactly the same.For this characteristic,this article uses a An algorithm based on phase-encoded image to divide the imaging area into multiple sub-areas according to the Point Spread Function(PSF)and Optical Transfer Function(OTF).Approximately considered to be invariant by wire.After the sub-region images are restored,multiple sub-region images are stitched into a complete high-resolution image.In addition,in order to further improve the restoration quality of the coded image,based on the non-linear space-invariant characteristics of the image and the idea of sub-region processing,this paper proposes a restoration algorithm based on BP neural network combined with image block.It solves the problem of inaccurate description of the degradation model of the traditional restoration algorithm,which greatly improves the accuracy of the degradation model.In summary,the main research focuses of this article include:(1)Analyze the image characteristics of different phase encoding functions according to the phase-encoding imaging principle to ensure the universality of the restoration algorithm in this paper;analyze the off-axis points and off-focus optical transfer functions to prove the nonlinearity of the actual optical system The invariant system provides a theoretical basis for iso-halo area division,system calibration method,and extraction of point spread functions.(2)Analyze the noise and ringing effect of the image based on the characteristics of the encoded image and the PSF distribution obtained from the sampling surface of the phase-encoded imaging to find out how to suppress the noise and ringing effect.(3)Since the actual optical system is a non-linear space-invariant system,different off-focus directions and different off-axis points on the sampling plane should correspond to different point spread functions.According to this characteristic,the sampled image is divided into equal halo regions.In this region,it is approximately considered to be linear and invariant.Therefore,according to the PSF corresponding to diifferent subregions,each subregion is restored first and then spliced into a height.Quality reconstructed image.(4)In the actual simulation and experiment process,the matching between halo areas such as images and the best PSF is difficult to achieve.It is often necessary to obtain hundreds of thousands of samples through a large number of calibrations and build a huge database at the same time.The matching relationship between the two is determined by the optimization algorithm,which makes the algorithm complicated and it is difficult to achieve efficient image restoration.In order to solve this problem,this paper uses BP neural network to calibrate the phase-coded diffraction imaging system in advance to construct the connection between the iso-halo region and the point spread function.In the later image restoration,only the iso-halo area is input into the network,and the optimal point spread function of the iso-halo area at this time can be obtained.In addition,the ant colony algorithm is used to optimize the training speed and accuracy of the BP neural network.