Michael Halle
Surgical Planning Laboratory
Department of Radiology
Brigham and Women's Hospital
Boston, Massachusetts, USA
mhalle@bwh.harvard.edu
Published in Computer Graphics, ACM SIGGRAPH, 31(2), May 1997, pp. 58-62.
This paper is also available as a PDF file.
The information contained in this paper was collected in part while the author was at the MIT Media Laboratory.
Copyright 1997 Michael Halle. Document last modified January 17, 1997.
Autostereoscopic displays present a three-dimensional image to a
viewer without the need for glasses or other encumbering viewing
aids. Three classes of autostereoscopic displays are described:
re-imaging displays, volumetric displays, and parallax displays.
Re-imaging displays reproject an existing three-dimensional object to
a new location or depth. Volumetric displays illuminate points in a
spatial volume. Parallax displays emit directionally-varying image
information into the viewing zone. Parallax displays are the most
common autostereoscopic displays and are most compatible with
computer graphics. Different display technologies of the three types
are described. Computer graphics techniques useful for
three-dimensional image generation are outlined.
After many years of relative obscurity, three-dimensional displays
have recently become both increasingly popular and practical in the
computer graphics community. This interest can be attributed to many
factors. In our daily lives we are surrounded by synthetic computer
graphic images in print and on television, and can now even generate
similar images on personal computers in our home. We also have
holograms on our credit cards and lenticular displays on our cereal
boxes. And has it really been so many years since we first saw
Princess Leia projected into thin air in the Star Wars motion
picture? In fact, the general public has been excited about
three-dimensional images since the days when stereoscopes graced
every mantelpiece at the turn of the century, through the 3D movie
craze of the early 1950's, the wonder of holography in the 1960's,
and the new frontier of virtual reality today. With each new
technology or movie, the excitement seems to grow.
Developments in the computer graphics industry have also done
their part to make spatial images more practical and accessible. In
the business of computer graphics, the computational power now exists
for desktop workstations to generate stereoscopic image pairs for
interactive display. At the high end of the computational power
spectrum, the same advances that permit intricate object databases to
be interactively manipulated and animated also permit large amounts
of image data to be rendered for high quality three-dimensional
displays. Finally, there seems to be a general realization in the
research and scientific community that the two-dimensional
projections of three-dimensional scenes traditionally referred to as
"three-dimensional computer graphics" are insufficient for
inspection, navigation, and comprehension of some types of
multivariate data. For these databases, the oft-neglected human depth
cues of stereopsis, motion parallax, and to a lesser extent ocular
accommodation are essential for image understanding.
The broad field of virtual reality has driven the computer and
optics industries to produce better stereoscopic helmet- or
boom-mounted displays, as well as the associated software and
hardware to render scenes at rates and qualities needed to produce
the illusion of reality. However, most journeys into virtual reality
are currently solitary and encumbered ones: user often wear helmets
or other devices that present the three-dimensional world to them,
and only to them. Presenting a three-dimensional image to a casual
passerby, a group of collaborators, or an audience requires a
different technology: autostereoscopic displays.
Autostereoscopic displays present a spatial image to a viewer
without the use of glasses, goggles, or other viewing aids.
Autostereoscopic displays are appealing because they offer the best
approximation to the optical characteristics of a real object. As a
result, though, there is much misunderstanding and misinformation by
those who would oversell the capabilities of a particular technology.
This paper will try to outline the strengths and practical
limitations of the different technologies by classifying them into
broad categories.
Our current understanding of physics does not include a practical
way of forcing photons to change direction in the absence of an
optical medium. Thus, a fundamental and general statement can be made
about all spatial displays, whatever its particular technology. This
paper will refer to this requirement as the projection
constraint:
Photons must originate in, or be redirected by, some material. The material can be behind, in front of, or within the space of the image, but it must be present. All claims to the contrary violate what we understand about the world. Figure 1 shows the possible relationships between the image and the display. A corollary to this constraint is the observation that air, water, or smoke are, in general, very poor display media. Images appearing "in mid-air", called aerial images, will invariably have originated not in the air from some other medium. Technologies lavished with claims of mid-air projection should always be scrutinized with regard to the fundamental laws of physics .
A specific and practical result of the projection constraint is that no matter where a spatial image appears with respect to its display, the image will be clipped by the display's physical boundaries. If for instance, an image appears in front of its display, a sufficient translation of the viewer will cause part or all of the object to intersect and "fall off" the edge of the display. This condition, known as a window violation, is particularly disturbing for aerial images.Figure 2 illustrates a window violation.
Physically realizable autostereoscopic displays can be classified
into three broad categories: re-imaging displays, volumetric
displays, and parallax displays. Re-imaging displays capture and
re-radiate the light from a three-dimensional object, perhaps to a
new location in space. Volumetric displays span a volume of space,
allowing individual parts of the space to be illuminated. Finally,
parallax displays are surfaces that radiate light of
directionally-varying intensity. Displays of each type have been used
in commercial display systems, and each has inherent strengths and
weaknesses.
Re-imaging displays are technically the simplest of
autostereoscopic displays. Re-imaging displays do not by themselves
produce a three-dimensional image. Rather, they affect the appearance
of another three-dimensional image in some way. The most trivial
re-imaging displays is a plain piece of glass. The back surface of
the glass intercepts rays of light traveling in different directions
with different intensities. The energy of the light is propagated to
the front surface of the glass, where light is re-radiated in the
same direction and with the same intensity as when it was captured.
Although the piece of glass is a very simple device, it does
illustrate that even a passive optical element can display an
arbitrarily complex light field that maintains the
three-dimensionality of a scene.
A mirror is only slightly more complex than plain glass, but it is
capable of altering the direction of all the rays of light entering
it without changing either the intensity or the color of the light
itself. A semi-silvered mirror can superimpose the light from two
three-dimensional scenes and re-radiate the result. Mirrors,
semi-silvered or not, are perhaps the most effective (and
cost-effective) three-dimensional displays used in theme parks
today.
More complex re-imaging displays are based on lenses and mirrors
with optical power that can translate the position of an object in
depth or distort it into a different three-dimensional shape. The
three-dimensional display by Dimensional Media Associates uses a
mirror system to relay a two- or three-dimensional object out in
front of the display surface [8]. If the two-dimensional object is a
CRT screen, a flat image of the screen will be appear to float in
front of the device. The location of the viewer must be restricted to
minimize window violations.
Another type of optical system was used by SEGA in an unsuccessful
arcade video game called "Time Traveler" (marketed under the
misleading term "hologram") to relay and distort the appearance of a
flat CRT into a curved surface. The successful commercial billing of
re-imaging displays as "holographic" systems is a clear statement
about how vivid the images they produce can be. On the other hand,
the use of these devices in technical display applications is
severely limited by their inability to display general
three-dimensional information. Optical re-imaging is often
incorporated into more general autostereoscopic displays.
Both of the other two classes of autostereoscopic displays
described here, volumetric displays and parallax displays, can
produce more general synthetic three-dimensional images. The major
difference between the volumetric displays and parallax displays lies
in the way they address the three-dimensional image volume. A
volumetric display addresses individual points in the volume
explicitly: input to this type of display can be a voxelized data
volume or a display list of three-dimensional primitives. This data
is then drawn in the three-dimensional space. In contrast, a parallax
display device images the direction and intensity of light at many
different locations on the display surface. Unlike the explicit
three-dimensional input for the volumetric display, the parallax
display's input consists of two-dimensional projections such as
photographic or synthetic images. Each two-dimensional image contains
no explicit depth information. Instead, depth is implicitly encoded
as positional disparity between different projections. The next two
sections look more closely at volumetric and parallax displays.
Volumetric displays work by filling or sweeping out a volume of space. In the inexact terminology of three-dimensional imaging, volumetric displays are also called volume displays, slice stacking displays, and space filling displays. Several technologies of volumetric displays exist. One of the most elegant is the varifocal mirror display [19], shown in Figure 3. A varifocal mirror display uses a mirrored membrane of varying optical power to sweep an image of a CRT through different depth planes of a volume. By synchronizing the CRT display with the mirror's oscillation, any point within the volume can be displayed. The greatest difficulty with varifocal mirror displays is building a high quality varifocal optic that can be oscillated at high frequencies.
Another group of displays use a spinning element to physically
sweep out a volume. The element can be a simple rectangle spinning in
a cylinder, or it can be a more complicated shape such as a helix.
The spinning element can either have light sources such as light
emitting diodes attached to it, or it can be scanned by an external
focused light source. An early example of the concept was patented by
Ketchpel in 1964 [9]. A example of a contemporary system is being
developed at the United States Naval Command, Control and Ocean
Surveillance Center [10]. Space-filling displays of this type will
always face the challenges of mechanical scanning.
A final type of volumetric display fills the volume with a display
medium that can be excited externally to the point where it emits
light. This external stimulus can, for instance, come from lasers of
different wavelengths that are scanned through the imaging medium.
Finding display materials with the appropriate non-linear optical
properties has proved to be a great research challenge. Recent
progress has been made at Stanford by Downing[3], incorporating
principles described by Lewis [11]. The ideal display material, not
yet discovered, must combine the qualities of optical efficiency, low
cost, and light weight to order to find widespread use.
Independent of the optical and mechanical technology, volumetric
displays share several properties. First, the image they present is
visible from a wide range of viewpoints, even permitting a viewer or
group of viewers to walk all the way around the display. Second, the
displayed image emits a continuous, uniform, spherical wavefront
centered at each displayed point. The human eye can selectively focus
on this wavefront, providing the sense of ocular accommodation. On
the other hand, because the wavefront is uniform and omnidirectional,
view-independent shading of objects is not possible. Even more
critically, current displays do not exhibit arbitrary occlusion of
one part of the image volume by another. For photorealistic scenes,
occlusion is almost always the most important depth cue, much more
important than ocular accommodation and often stronger even than
stereopsis. Volumetric displays thus have most common use in
non-photorealistic applications such as wireframe images and
icon-based displays. Volumetric devices are appropriate for this type
of application because they can vector-scan only the regions of space
that the object spans, eliminating the display bandwidth that would
otherwise be required to rasterize the entire image volume.
A similar technology to volumetric displays, stereolithography,
has achieved widespread use in the CAD/CAM industry because the
result of the scanning process is not the emission of light, but a
formation of a solid computer-generated object made from a polymer
material. Also related to volumetric displays is VOXEL Corporation's
VOXBOX holographic display system for radiology [5]. The VOXBOX uses
a hologram to display stacks of tomographic images as an image
volume. It shares many imaging properties of volumetric displays.
Parallax displays consist of a surface covered with display
elements that can emit light of varying intensity in different
directions. The plain piece of glass "display" is a good way of
thinking about parallax displays: the front surface of the glass is a
continuum of sites that send out a hemisphere of light varying in
intensity. A single output site radiates only information captured
from one viewpoint, the corresponding site on the back surface of the
glass. While this site's information contains no explicit notion of
depth, the light emitted from several sites considered as a whole
presents a three-dimensional image. Depending on a viewer's exact
location, light traveling in different directions appears to
originate from different parts of the glass. This visual information,
intercepted by the viewer's two eyes, is processed to form a
three-dimensional mental model of the scene.
The plain glass display analogy also demonstrates that parallax
display devices can correctly show arbitrary occlusion of one part of
an object by another. Occlusion is essential for the display and
comprehension of photorealistic synthetic scenes. On the other hand,
many types of parallax displays use information reduction techniques
that approximate the shape of the wavefront of light-emitting points.
These approximations diminish or eliminate any ocular accommodation
depth cues.
Holographic displays [1][17] are in many ways very close to the
"piece of glass" display model. Holograms store wavefront information
about an object as microscopic interference fringes during the
holographic exposure process. When the developed hologram is
illuminated, its interference fringe pattern acts as a complex
diffractive lens that reconstructs the object light's direction and
intensity. A holographic display reconstructs light so exactly that
it shares many of the properties of a volumetric display, including
providing ocular accommodation cues, without having to physically
span the imaging volume. Unfortunately, the data bandwidth of a high
quality display hologram is far beyond any current or envisioned
image synthesis technology: a typical display has a spatial frequency
exceeding 1500 line pairs per millimeter. In order for a hologram to
be synthesized, the information that is contains must be reduced to
manageable quantities.
Unfortunately, the alluring and eerie realism of holographic
displays has lead to an increasingly common misuse of the term
"hologram" to describe any display that is vaguely three-dimensional,
and even some that are not. To be clear, display holograms are
image-bearing diffractive optical devices. Other displays may be
three-dimensional, but they are not holograms.
Several parallax display technologies were developed long before
holography. The earliest were the parallax stereogram [6] and the
parallax panoramagram [7]. Both display types depend on a device
called a parallax barrier, an opaque material slotted with a series
of regularly spaced vertical slits. A piece of film or other imaging
medium is offset some distance behind the parallax barrier. Each slit
in the barrier acts as a window onto a stripe of the section of film
that lies behind it. Exactly which stripe of film is visible depends
on the horizontal angle from which the slit is viewed. A parallax
stereogram displays a stereoscopic image pair by interleaving columns
of the two images on the film, one column of each image per slit. An
appropriately-positioned viewer will see the right view of the pair
through the slits with the right eye, the left view with the left.
The parallax barrier blocks the opposite image from view. A
stereoscopic image is thus produced.
The parallax panoramagram extends this concept by introducing thinner columns of more views behind each slit. Artn's PSCHologram is an example of this technology [16]. The more views that are present, the more naturally the image will appear to change as the viewer moves from side to side. Figure 4 shows a parallax panoramagram. Parallax panoramagrams are limited because the barrier, while necessary to block the unwanted views, also blocks light from getting to the viewer. Panoramagrams usually require banks of bright, diffuse lights located behind the film. Displays of this type rely on the fact the spatial and directional information is spatially multiplexed onto the film, which leads to several other problems. First, a viewer positioned far enough to the left or right of the display will be able to look through one slit to see the image data associated with the slit's neighbor. As a result, the image appears to repeat its perspectives as the viewer moves. If the viewer sees a correct view with one eye and repeated view with the other, the depth of the object can even appear to flip inside out (called a psuedoscopic image). Second, the resolution of the film limits the maximum number of views that can be displayed. The spacing of the slits determines the maximum spatial resolution of the display.
The parallax panoramagram is three-dimensional only in the
horizontal direction; vertically, the image of the display behaves
just as if it were a flat photograph. As a viewer moves closer or
further from the display, vertical edges of the image will appear to
shift naturally with respect to each other, just as they would in a
real object. Horizontal edges, though, remain fixed relative to each
other. This kind of display is said to be horizontal parallax only,
or HPO. HPO displays are a useful engineering trade-off because they
greatly reduce the information content of a three-dimensional image
while still displaying stereoscopic and motion parallax information.
For demanding applications, the view limitations and possible
distortions of HPO would preclude its use in favor of a full parallax
display.
The word "lenticule" is a synonym for "lens", but the term
"lenticular" has come to refer to a type of three-dimensional display
that using an array long, narrow lenses instead of slits to display
three-dimensional information. More correctly, this display type
should be referred to as a lenticular panoramagram. Figure 5 depicts
a lenticular panoramagram.
This display type is functionally very similar to the parallax
panoramagram. Each lens focuses on the image information located
behind it and directs the light in different directions. If we think
of one slit of a parallax panoramagram as similar to a camera with a
pinhole as an aperture, one lenslet of a lenticular panoramagram is
analogous to a camera lens. Cameras with lenses are more common than
pinhole cameras because they collect more light from the scene.
Similarly, a lenticular panoramagram is brighter and optically more
efficient than the corresponding parallax panoramagram. The entire
surface of the lenticular sheet radiates light; there are no dark
stripes such as those produced by a parallax barrier.
Continuing the camera analogy, camera lenses come in a wide
variety of focal lengths and materials and can be adjusted for focus,
while the only adjustable variables of a pinhole camera are the
pinhole diameter and the spacing between the pinhole and the film.
Lenticular sheets are molded from plastic in a process that sets the
width of the lenslets, the distance between the image and the lens,
and each lenses' optical power. The quality of the sheet-making
process also determines the optical aberrations that will be
manifested in the final image. The optical power of the lens controls
the angle of view through which the final image can be seen.
Lenticulars are almost always made so that the film plane is located
one focal length behind the lenses: the image data emerges collimated
from each lenslet. Making high quality yet affordable lenticular
sheets is one of the major difficulties of creating lenticular sheet
displays.
Lenticular panoramagrams can also be used with a CRT or other
two-dimensional display device to produce a dynamic three-dimensional
image. The spatial resolution of the two-dimensional display must be
high enough in the horizontal direction to provide both spatial and
directional information. Optical alignment of the underlying display
with the lens sheet is essential to producing distortion-free
three-dimensional image. As always, the information bandwidth of the
display increases as more directional information is added.
Like parallax panoramagrams, lenticular panoramagrams display only horizontal parallax. Another display type, the integral photograph or integram, uses spherical lenses instead of cylindrical ones to present horizontally and vertically varying directional information, thus producing a full parallax image. Figure 6 shows the integram's spherical lens array. Integrams are less common than their cylindrical lensed counterparts mostly because even more of their spatial resolution is sacrificed to directional information.
The optical systems used in parallax and lenticular panoramagrams
are tightly constrained by their how much information can be stored
on their imaging medium. They are also prone to image repeating
caused by crosstalk between adjacent display elements. The
holographic stereogram overcomes these problems by combining the
information storage capacity of the hologram with the
information-reducing image discretization of the panoramagram [2].
(As Okoshi notes, the holographic stereogram should rightly have been
called a holographic panoramagram [15].) Instead of macroscopically
encoding directional information as interleaved stripes of image, a
holographic stereogram optically records the same information
microscopically as fringe patterns.
The details of the holographic stereogram's recording process are
beyond the scope of this text. The basic idea of the process is to
make a series of holographic windows or aperture through which
two-dimensional projections of image data can be seen. Each aperture
is a distinct optical element; this property eliminates the problem
of element crosstalk and image repeating. "One-step" holographic
stereograms are recorded directly onto the final display media using
spatially discrete slit apertures. One-step holograms are useful for
rapidly produced single-copy images. "Two-step" holographic
stereograms use a spatially discretized master hologram to record the
image information. The master hologram can be transferred to make
many final holographic images. The two-step process is most
frequently used when multiple copies of the hologram are desired.
Holographic stereograms can be recorded in a number of media and
are suitable for stamping and publication. The holographic
stereogram's major weakness is the complexity of its exposure
apparatus and the difficulties of lighting and color that result from
the hologram's diffractive properties.
The fact that display holograms are even possible is directly due
to the fact that the physical process of interference that forms the
holographic fringe pattern is a parallel optical computation.
Electro-holography, the technology of forming the same type of fringe
pattern electrically, does not have the benefit of this natural
computer. Instead, the fringe patterns for electronically-generated
holograms must be conventionally computed and then output to a
physical device capable of diffracting light and producing the
three-dimensional image. Electro-holography depends heavily on
information reduction techniques such as the elimination of vertical
parallax and holographic stereogram-like discretization of spatial
and directional information. Recent progress on developing both the
display technology and the computational algorithms used to compute
the fringe patterns has been encouraging [18] [12], but the
technology is still far from producing high quality three-dimensional
images using affordable hardware.
At the current time, parallax displays are the three-dimensional
displays most commonly used with computer-generated images. Parallax
displays are popular for three primary reasons: they can be published
and mass-produced, they can be made in a wide range of sizes, and
they can produce photorealistic images. For these reasons, these
basic guidelines for generating effective spatial images will
concentrate on parallax displays. Here are some important things to
consider when undertaking the process of creating three-dimensional
displays.
Stereopsis is only one of the cues we use to evaluate an object's
dimensionality. Without stimulation and agreement of all depth cues,
spatial images can appear uninteresting or even visually confusing.
Strong occlusion, shading, and perspective depth cues are even more
important in three-dimensional images than in 2D renderings and
photographs.
Some day, software packages will take care of the details of
matching rendering to a display's characteristics, allowing the
designer to concentrate on the content of the image. Unfortunately,
that day is not yet here. Each display type has its own properties,
and each manufacturer has their own particular process. These
constraints simply must be respected in order to produce undistorted,
high quality images.
If you are involved in your first three-dimensional imaging
project, it is best to work either directly with the manufacturer or
with a designer who has experience creating such images. They will
able to help you plan your imagery to avoid common problems such as
window violations and conflicting depth cues. They will most often
also be able to provide correct parameters for a particular
display.
The process of rendering for most parallax displays consists of moving a computer graphics camera along a track in front of the virtual scene, capturing images at regular intervals. The varying camera viewpoints provide the spatial and directional information that is encoded into the display when it is made. Here are some basic camera and image parameters required for generating parallax displays:
Horizontal parallax only images have a specific viewing distance
where the varying horizontal and fixed vertical perspectives of the
object match. Viewers not located at the correct distance will see
images with cylindrical lens distortion. To minimize this problem, be
sure to match the computer graphic camera's view distance to that
specified for the display.
In addition, all parallax displays impose inherent limits on the
spatial resolution of the image volume that depends on image depth
[4]. For example, the horizontal resolution of a parallax
panoramagram is limited by both the barrier slit width and the width
of the vertical image slices on the film. Similarly, the horizontal
spatial resolution of a lenticular panoramagram cannot exceed the
width of a lenslet at the display surface. Lenticular sheets are also
limited by lens blur and diffraction. Exceeding these limits produces
images that break up into horizontal, jaggy pieces. Many
manufacturers may be only anecdotally familiar with these
limitations.
If you are trying to get the best possible spatial image, you must
go beyond these basic guidelines and actually model the imaging
process of image synthesis, recording, and display. The computer
graphics image data must match where the light from the final display
actually goes. One of the most difficult part of this process is not
the computer graphics algorithms, but actually measuring and
understanding the optical characteristics, geometry and distortions
of the display device. Correct matching of graphics and display can
produce significant gains in image realism and lucidity, but
simplifying this process remains an area of continuing research and
development.
A range of autostereoscopic displays exist today that can be used
in applications ranging from advertising tools to air traffic control
consoles and publishable images for scientific visualization. As the
demand for such displays increases, the 3D and computer graphics
fields face the challenge of demystifying three-dimensional
technology and simplifying the image generation process. Sensible
image design, selection of an appropriate display device, and
adherence to its limitations can yield realistic, understandable, and
uniquely effective three-dimensional images
[1] Benton, Stephen A. Display Holography: an SPIE Critical Review
of Technology. Proc. SPIE, Holography, A86-32351 14-35, (1985), pp.
8-13.
[2] Benton, Stephen A. Survey of Holographic Stereograms. Proc. SPIE,
Processing and Display of Three-Dimensional Data, 367, (1982), pp.
15-19.
[3] Downing, Elizabeth et. al. A Three-Color, Solid-State,
Three-Dimensional Display. Science 273, 5279 (August 30, 1996), pp.
1185-1189
[4] Halle, Michael W. Holographic Stereograms as Discrete Imaging
Systems. Proc. SPIE, Practical Holography VIII, 2176, (May 1994), pp.
73-84.
[5] Hart S.J. and M.N. Dalton, Display Holography for Medical
Tomography, Proc. SPIE, Practical Holography IV, 1212 (May 1990), pp.
116-135.
[6] Ives, F. E. U.S. Patent No. 725,567, (1903).
[7] Ives, H. E., A Camera for Making Parallax Panoramagrams, J. Opt.
Soc. Amer., 17, (Dec. 1928), pp. 435-439.
[8] Kawamoto, Wayne 3-D Images That Float in Air. Byte (Oct.
1995).
[9] Ketchpel R. D. U.S. Patent No. 3,140,415, (July 7, 1964).
[10] Lasher, Mark et. al. Laser-Projected 3D Volumetric Displays.
Proc. SPIE, Projection Displays II, 2650 (1996), pp.285-295.
[11] Lewis, J. D. et al. A True Three-Dimensional Display. IEEE
Transactions on Electron Devices, ED-18, 9, (Sept. 1971), pg.
724-732.
[12] Lucente, Mark and Tinsley A. Galyean. Rendering Interactive
Holographic Images. Proceedings of SIGGRAPH '95 (Los Angeles, CA,
Aug. 6-11, 1995). In Computer Graphics Pro ceedings, Annual
Conference Series, 1995, ACM SIGGRAPH, New York, 1995, pp
387-394.
[13] McAllister, David F. Stereo Computer Graphics and Other True 3D
Technologies. Princeton Univ. Press, 1993.
[14] McKenna, Michael and David Zeltzer. Three Dimensional Visual
Display Systems for Virtual Environments. Presence, 1, 4 (Fall 1992),
pp 421-458.
[15] Okoshi, Takanori. Three-Dimensional Imaging Techniques. Academic
Press, New York, 1976.
[16] Sandin, Daniel J., et. al. Computer-Generated Barrier-Strip
Autostereography. Proc SPIE, Non-Holographic True 3D Display
Technologies, 1083, (Jan. 1989), pp. 65-75.
[17] Saxby, Graham. Practical Holography, 2nd edition. Prentice Hall,
Dec. 1994.
[18] St.-Hilaire, Pierre et. al. Scaling Up the MIT Holographic Video
System. Proc. SPIE, Fifth International Symposium on Display
Holography, 2333, (Feb 1995), pp. 374-380.
[19] Traub, A.C. Stereoscopic Display Using Varifocal Mirror
Oscillations. Applied Optics, 6,6 (June 1967), pp. 1085-1087.