Signal Processing - September 2016 - 126

patterns that result in a sharp target image when observed with a
With the re-emergence of near-eye displays for virtual realdefocused eye to be well-posed or invertible. One could attempt
ity (VR) and augmented reality (AR), focus cues have recently
to preprocess an image for a 2-D display, such that when the eye
attracted a lot of attention in these applications. Most existing
is accommodated away from the physical screen it looks like
near-eye displays naturally provide stereoscopic cues because
the target image [19]. However, it turns out that this approach is
either two separate microdisplays provide an image for each eye
ill-posed and therefore not invertible. But even when the image
or a single screen is optically split with two lenses. The design
formation is invertible, vision-correcting disprinciples of near-eye displays available
plays require the prescription of the viewer
today are very similar to the stereoscopes
With an ever-increasing
to be known (no changes to the hardware are
widely used in Victorian times [16]. In this
demand on image
necessary, different prescriptions can be corcontext, light field displays offer the possiresolution, one of the
rected in software) and the pupil positions
bility of providing focus cues. The authors of
major bottlenecks in the
and diameters to be fixed or tracked. Addi[17] were the first to demonstrate an integrallight field display pipeline tionally, this technology has not yet been
imaging-type near-eye display that allowed
shown to simultaneously correct for visual
for a wide accommodation range of the
is computation.
aberrations and provide stereoscopic cues in
observer with the common caveat of reduced
direct-view displays; however, near-eye light field displays,
display resolution. (The range was wide enough that correction
which employ a separate display per eye, have been demonstratof refractive errors was supported, allowing the user's prescriped to simultaneously correct visual aberrations while presenting
tion eyewear to be removed and digitally correctly by altering
binocular images [17].
the imagery depicted by the near-eye light field display.) More
recently, [13] investigated compressive light-field synthesis via
two stacked LCDs. This is shown in Figure 6. The device design
High-dynamic-range displays
is inspired by common stereoscopes but it employs two LCD
The dynamic range of a display usually refers to its contrast,
panels spaced at about 1 cm in the display housing. Using light
i.e., the ratio of brightest possible display state and the darkfield factorization algorithms similar to those employed by the
est. The dominant display technology for achieving an HDR
content-adaptive parallax barriers described previously, a 4-D
is a combination of low-resolution light emitting diodes
light-field is emitted independently to each eye, providing paral(LEDs) and a high-resolution LCD [5]. The rationale for this
lax over the eye box.
design is that the LEDs can be turned off completely, thereby
producing a truly black image, or they can be boosted to produce extremely high peak brightness. The main challenge is
Vision-correcting light field displays
that LED arrays cannot yet be easily produced at a high resoPerhaps one of the most unconventional applications of light
lution and small pixel size. Thus, the HDR display uses a lowfield displays is correction of visual aberrations for a human
contrast, but high-resolution LCD to deliver high image
observer [12], [17], [18]. Instead of correcting vision with eyeresolution while the low-resolution, high-contrast LED array
glasses or contact lenses, the same can potentially be done
serves as a programmable backlight. Just over a decade after
directly in the screen, allowing for myopia, hyperopia, astigmapublication, HDR displays are now widely available in the
tism, and even higher-order aberrations to be corrected. For
consumer market, usually under the name "microdimming"
such an application, the light field display presents a distorted
or "local dimming."
light field to the eyes of the viewer such that their natural aberIn addition to providing degrees of freedom for glassesrations optically undistort the light rays, resulting in the desired
free 3-D image synthesis, stacked LCDs also decrease the
image (Figure 7). This idea is somewhat similar to wavefront
black level of the produced image compared to any individual
correction with adaptive optics. Implemented with light field
LCD. Although the peak brightness of a multilayer display is
displays, the requirement on the angular resolution (density of
usually reduced compared to each of the respective layers, the
emitted viewing zones) is similar to those for light field displays
observed contrast is increased. This makes multilayer displays
supporting focus cues: multiple different viewing zones have to
ideal display systems for HDR image generation. However,
be displayed into the same pupil. Assuming that the prescription
with nonnegligible separations between attenuators, multilayof a viewer is known and that their pupil location and diameter
er HDR decomposition becomes a 3-D display problem simican be tracked by a camera, light field displays attempting
lar to light field image synthesis, since all viewpoints must
vision-correction constrain the light field synthesis to the pupil
produce an accurate rendition of the 2-D image within the
locations. For example, simple implementations using parallax
target field of view. The factorization framework described by
barriers were shown by [12] and, using integral imaging, by
[9], [10] inherently accounts for the limited contrast of each
[18] and [17]. As shown by Huang et al., the pupils are tracked
layer, thereby allowing simultaneous optimization of dynamic
and the light field is dynamically rendered and predistorted for
range and accurate multiview imagery. In a 2-D HDR disthe viewer such that only viewing zones that actually enter the
play mode, the target light field encodes a single plane (e.g.,
pupil are considered. Using a 4-D light field display for vision
coincident with the front layer), with a texture given by the
correction, as opposed to a conventional 2-D display, has the
desired HDR image. Figure 8 shows the result from a parallaxadvantage that the additional degrees of freedom afforded by the
free 2-D HDR display prototype. The optimized layers in (b)
four dimensions allow for the inverse problem of synthesizing
126

IEEE SIgnal ProcESSIng MagazInE

September 2016

Table of Contents for the Digital Edition of Signal Processing - September 2016