Research On Image-preprocessing-based Aberration Optimization Methods For 3D Light-field Display

The technology revolution continues to evolve rapidly,and a new round of industrial transformation has entered into a critical period,right here right now.As a cross-disciplinary technology that integrates optics,material science,computer science,electronic engineering,communication technology and so on,three-dimensional(3D)display technology has carried humans’ beautiful yearning for a higher-quality,hyper-realistic next generation display.Also,the Outline of the 14th Five-Year Plan(2021-2025)for National Economic and Social Development and Vision 2035 said that the development of equipment and solutions for virtual reality should be taken as a national-duty-call.Among the existing 3D display technologies,the 3D light-field display can reconstruct the light distribution of the target scenes,to present the spatial information correctly.However,3D light-field display technology remains a worldwide cutting-edge technology,there are still many issues such as limited viewing angles,low viewpoint density,serious cross-talk,low image quality,and insufficient depth of field that need to be solved.This work studies the aberration optimization methods for 3D light-field display aiming to improve the displayed images’ quality in the 3D light-field display with large viewing angles.The contents of this work are mainly as followed:(1)Aberration correction based on a pre-correction convolutional neural network for 3D light-field displaysTo enhance the image quality while not decrease the viewing angle of the integrated imaging 3D light-field display,In Chapter 3,this work analyzed the existing optical aberration correction methods of 3D light-field display,and designed a lens to improve the viewing angle of the display system,then,proposed a pre-correction method based on a neural network(CNN)for 3D light-field display with large viewing angle.The proposed method modeled the imaging process of the lens mathematically and added it to the CNN’s training procedure,then pre-corrected the elemental images with the trained model,so as to improve the imaging quality without adding new hardware or reducing the viewing angle.Also,a light-field display prototype is built based on the proposed method.In the experiment,a virtual camera array was used to capture the 3D scene,and pre-corrected the elemental images with the trained model,then a 3D image is reconstructed with the 3D light-field display prototype.A full-parallax 3D scene with a 70-degree viewing angle is presented to the viewers.(2)Regional selection-based pre-correction of lens aberrations for 3D light-field displaysIn a lens-based integrated imaging 3D light-field display using circular lenses as the primary light-control device,the severity of aberration varies between different field of view(FoV)under a lens.Among the existing aberration pre-correction methods,using a Wiener filter can produce ringing artifacts in fields with severe aberrations,while the corrected results of the CNN are poorly(not as good as pre-corrected by Wiener filter)in fields with slight aberrations.For solving these problems,in Chapter 4,this work proposed a regional selection-based pre-correction method based on the severity of the aberration in different fields.That is,the CNN is applied to pre-correct the "regions with severe aberration",while the Wiener filter is applied to pre-correct the "regions with slight aberration".The simulation and experimental results proved that after applying the proposed region-selective pre-correction algorithm to those elemental images,the overall quality of the reconstructed 3D images had improved.