The emerging light field technology is playing an increasingly significant role in computational imaging and computational display. Consumer light field camera has demonstrated attractive photographic applications like refocusing, viewpoint changing, and synthetic aperture. In the 3D display field, light field has been regarded as one of the most promising techniques to produce realistic 3D visual effects and solve the vergence-accommodation conflict problem. This thesis explores light field technology in two aspects: light field synthesis and light field display.

A light field camera captures both the spatial and directional information of light rays coming into the camera in a single exposure. However, there is an inherent trade-off between the spatial and angular resolutions as the spatial pixels are divided into many angular views. In addition, specialized device is needed for capturing, which further hinders the popularization of light field technology. If a 4D light field can be synthesized directly from 2D RGB image, we can bypass the light field capturing and avoid the spatial and angular resolution trade-off. To this end, we propose a deep neural network called Light Field GAN (LFGAN) that synthesizes a 4D light field from a single 2D RGB image directly. We introduce novel loss functions including epipolar plane image (EPI) loss and brightness loss to assist the network in learning the geometry and color information. The synthesizing performance are discussed and applications such as refocusing and viewpoint changing are presented.

Different from 2D image, the synthesized 4D light field cannot be displayed on traditional display panel. Therefore, we introduce a method to create light field display using a single projector and an array of plane mirrors. Mirrors are used to reproduce densely arranged virtual projectors regardless of the physical size of the real projector. Compared with traditional projector-array-based display system, the proposed mirror-array-based display system produces much higher ray density under much lower budget. Such high-density light field allows more than one ray depicting the same 3D point enters the pupil of the viewer's eye at slightly different angles simultaneously, thus naturally driving the eye's lens to focus at the depicted depth rather than the screen surface and thereby effectively alleviating the vergence-accommodation conflict, which is a very common problem of conventional stereoscopic and multiview 3D displays. We propose an ellipsoidal geometric framework and a design pipeline, and use parametric modelling technique to automatically generate the display configurations satisfying target design parameters.