High quality computer generated hologram computation and display applications

Computer Generated Holographic (CGH) displays can be the ultimate three-dimensional (3D) display technology as they can match all the requirements of the human visual system. A CGH display goes beyond stereoscopic displays and offers a 3D experience with all the depth cues, including vergence and accommodation. CGH displays can satisfy all the demands of industrial, educational, and consumer applications. Despite its great potential, the visual quality of CGH displays is inferior to other display technologies. There remain many computational and implementation challenges that are needed to be resolved before fulfilling its promises. In this thesis, we developed two novel methods to improve the computation and the visual quality of holograms. First, we explored a new direction called learned holographic light transport models, where we improved hologram computation using machine learning approaches. Although traditional computational models offer excellent quality in simulation environments, experimental images from these models do not exhibit the expected visual quality. Our work addresses this mismatch by generating a holographic dataset and learning model that contains the reconstructed images for simulated and experimental results. We proved that our method mitigates the mismatch between simulated and experimental results while improving the visual quality of experimental results. The second challenge is the mismatch between the incoherent natural scenes and the holographically constructed scenes using coherent light such as lasers. While the blur due to defocus results in smoothed features in natural images, defocus blur in coherent images contains high-spatial frequency features, which can inadvertently disturbs the eye's accommodation in 3D holographic scenes. Such difference in defocus blur manifests itself as edge fringe artifacts in 3D holography. We investigated these artifacts, and proposed a novel phase only hologram generation method to mitigate this issue. Our method introduces a novel targeting scheme and loss function that is specifically tailored to improve the visual quality by reducing these artifacts. Furthermore, we propose a new optimization method we named the Dual Stochastic Gradient Descent method. Defocus blur is a well-known problem in coherent systems, but previous research has failed to show any enhancement. We showed for the first time that our method could reduce the edge-fringe artifacts both in simulations and experimental results. Finally, we developed a novel holographic vision simulator device to assess the post-visual acuity performance of cataract patients before going through surgery. The device contains a CGH display and pupil tracker cameras. We demonstrate that holographically shaped light beams can be programmed and directed through less dense cataractous regions of the crystalline lens to form crisp images on the retina. Our pre-clinical studies with 13 patients showed that patients' potential post-op visual acuity after surgery can be successfully predicted using the CGH display before the surgery. Such a simulator has enormous potential in the clinic.