High-quality integral imaging 3D display from a captured monocular image

Due to the limited space-bandwidth product, three-dimensional (3D) integral imaging (InI) encounters the trade-off between spatial resolution and viewpoint density. This problem significantly impedes the advancement of high quality InI 3D displays. Moreover, the cost and complexity of light field sampling constrain the generation of 3D light fields for real scenes. Here, we propose a high-quality light field rendering pipeline utilizing a captured monocular image. The proposed elemental image array (EIA) generation pipeline consists of a depth generator (DG), forward light tracing (FLT), and an elemental image super-resolver (EISR). The DG predicts the depth map of the captured monocular image. FLT synthesizes the monocular image and depth map to produce the EIA. The EISR optimizes the generated EIA, enhancing the spatial resolution of the reconstructed 3D scene. Our method facilitates the miniaturization and integration of light field sampling devices. Compared with conventional methods, the average structure similarity of the predicted depth map increases by 54%, and the average peak signal-to-noise ratio of the reconstructed image improves by 50%. Optical experiments demonstrate that the proposed method implemented on our InI prototype achieves high-quality light field 3D display for both virtual and real scenes using a streamlined sampling pipeline.