Research On Problems Of Spatial Information Sampling Rate And Scattering Imaging In Micro-lens Array-based Light Field Imaging Technology

Micro-lens array-based light field imaging is a computational imaging technology,which uses micro-lens array to record the light field information of the target scene(including the spatial information and the directional information of incident light).3D reconstruction of the object can be realized through the processing of light field information.Micro-lens array-based light field imaging has a compact system,can be conducted without the requirement of special illumination and has the high-temporalresolution 3D imaging feature.It has shown great value in biomedical,industrial and other application fields.Improving the imaging capability of micro-lens array-based light field imaging technology and satisfying the requirements of different application scenes are the focuses of relevant research,which require improving the information capturing capability of the imaging system and considering the influence of specific application scenes on the imaging process.Therefore,the problem of low spatial sampling rate of light field information and the problem of light field imaging in scattering scenes,which are important application scenes,have become two key problems.This dissertation focuses on the research of micro-lens array-based light field imaging,aiming at improving its imaging capability,and carries out research on the above two key problems.The dissertation includes the following works:1.In order to solve the problem of low spatial sampling rate,the spatial sampling rate increasing method by rotating the imaging object is proposed.3D reconstruction is conducted by using digital refocusing and 3D deconvolution with light field information with high spatial sampling rate.The grid-like artifacts in the refocusing images are eliminated,and the quality of the refocusing images are improved.On this basis,the application research of light field imaging in 3D topography measurement of micro-structured surface is carried out,and the height measurement precision of the workpiece is improved.2.In the scattering imaging scenes,when the effect of scattering on light propagation is unknown,the light field imaging forward model cannot be built.To solve this problem,the light field imaging through scattering medium with unknown characteristics based on 3D blind deconvolution is proposed.The scattering light propagation model based on Gaussian function polynomial is used to build the light field imaging forward model of scattering scene.The 3D reconstruction and the scattering forward model are optimized synchronously by alternate update in iterations.3.Scattering will introduce blur into the light field information and enhanced the degree of ill condition of the inverse problem of 3D reconstruction.Then the quality of3 D reconstruction will degrade.To deal with this case,a deep learning-based light field imaging method through scattering medium is proposed.The scattering of light field image will be removed by the neural network,and then the high-quality 3D reconstruction will be realized by using scattering-free forward model and the scattering-removed light field image.In addition,a neural network training method based on simulation samples is proposed to reduce the difficulty of obtaining training samples in 3D scattering imaging scenes.