Efficient Acquisition, Representation, and Rendering of Light Fields

Abstract

In this thesis we discuss the representation of three-dimensional scenes using image data (image-based rendering), and more precisely the so-called light field approach. We start with an up-to-date survey on previous work in this young field of research. Then we propose a light field representation based on image data and additional per-pixel depth values. This enables us to reconstruct arbitrary views of the scene in an efficient way and with high quality. Furthermore, we can use the same representation to determine optimal reference views during the acquisition of a light field. We further present the so-called free form parameterization, which allows for a relatively free placement of reference views. Finally, we demonstrate a prototype of the Lumi-Shelf system, which acquires, transmits, and renders the light field of a dynamic scene at multiple frames per second.