This paper presents a study of snapshot-compressed light field (LF) imaging, which reconstructs the LF, represented as a set of images taken from densely spaced viewpoints, from a single image (i.e., a snapshot). While conventional techniques have been attempted such as coded aperture and focal stack, which use multiple images to reconstruct LF, this paper focuses on two methods that extend these techniques to snapshot-compressed LF imaging: time-multiplexed coded aperture (TMCA) and coded focal stack (CFS). These methods embed observations at different apertures and focal depths for each pixel, allowing the necessary information for LF reconstruction to be embedded in a single image.

This study develops a deep learning-based pipeline to reconstruct the LF from images obtained using both methods, and compares and evaluates it with other LF reconstruction methods. The results show that TMCA and CFS achieve high reconstruction quality compared to other snapshot LF reconstruction methods, and exhibit performance that is reasonably good compared to methods that use multiple images.

The main contribution of this work is to evaluate and demonstrate the superiority of TMCA and CFS under comparable conditions for snapshot-compressed LF imaging. In addition, the development of a deep learning framework that unifies both methods is highly appreciated. The comparison and verification of the effectiveness of both methods in the context of snapshot LF imaging is the first attempt of its kind and is a scientifically significant achievement towards achieving efficiency and quality in LF imaging.