Efficient Rendering and Compression for Full-Parallax Computer-Generated Holographic Stereograms.

PhD thesis, Cornell University, 2000.

In the past decade, we have witnessed a quantum leap in rendering technology and a simultaneous increase in usage of computer generated images. Despite the advances made thus far, we are faced with an ever increasing desire for technology which can provide a more realistic, more immersive experience. One fledgeling technology which shows great promise is the electronic holographic display. Holograms are capable of producing a fully three-dimensional image, exhibiting all the depth cues of a real scene, including motion parallax, binocular disparity, and focal effects. Furthermore, they can be viewed simultaneously by any number of users, without the aid of special headgear or position trackers. However, to date, they have been limited in use because of their computational intractability. This thesis deals with the complex task of computing a hologram for use with such a device. Specifically, we will focus on one particular type of hologram: the holographic stereogram. A holographic stereogram is created by generating a large set of two-dimensional images of a scene as seen from multiple camera points, and then converting them to a holographic interference pattern. It is closely related to the light fields or lumigraphs used in image-based rendering. Most previous algorithms have treated the problem of rendering these images as independent computations, ignoring a great deal of coherency which could be used to our advantage. We present a new computationally efficient algorithm which operates on the image set as a whole, rather than on its individual elements. Scene polygons are mapped by perspective projection into a four-dimensional space, where they are scan-converted into 4D color and depth buffers. We use a set of very simple data structures and basic operations to form an algorithm which will lend itself well to future hardware implementation, so as to drive a real-time holographic display. We also examined issues related to the compression of stereograms. Holograms contain enormous amounts of data, which make storage and transmission cumbersome. We have derived new methods for efficiently compressing this data. Results compare favorably with existing techniques. Finally, we describe an algorithm for simulating a camera viewing a computed hologram from arbitrary positions. It uses wave optics to track the propagation of light from the hologram, through a lens, and onto a film plane. This enabled us to evaluate our rendering and compression methods in the absence of an electronic holographic display and without the lengthy processing time of hardcopy holographic printing.

This paper is available as a PDF file Kar00.pdf (8.4M).