Signal Processing - September 2016 - 125

Left View

Right View

LCD Patters at Frame 1

LCD Patters at Frame N

(a)

(b)

Figure 5. A compressive light-field prototype. The prototype uses three stacked layers of liquid crystal displays that are rear-illuminated by a single backlight. A light-field factorization algorithm computes time-multiplexed patterns for all LCD layers that are displayed at a speed exceeding the critical flicker
fusion threshold of the human visual system. Perceptually, these patterns fuse into a consistent, high-resolution light field that supports stereo cues and
parallax without the need for glasses.

In particular, multiplexing methods (e.g., temporal, spatial, polarization, etc.) can be adopted to optimize the tradeoff
between spatial and angular resolution, brightness, etc. in a content-adaptive manner. For example, the refresh rate of modern
displays often exceeds the critical flicker fusion (CFF) threshold
of human vision. A parallax barrier display implemented with
fast LCD panels would allow for the optimal layout of time-multiplexed pinholes to be determined for each target light field. Further relaxing the requirement that the barrier mask is constrained
to showing only pinholes leads to the concept of content-adaptive parallax barriers that optimize the time-multiplexed patterns
for both display and barrier mask [8]. Such a content-adaptive
optimization not only allows adaptive tradeoffs between spatial
and angular resolution to be made, but it also allows for display
brightness to be optimized with respect to pinhole-based barriers. The light field ul (x, y) generated by a time-multiplexed,
content-adaptive parallax barrier with two LCDs is given by (1).
The corresponding inverse problem is usually formulated as a
numerical optimization problem, which can be efficiently solved
with NMF approaches.
Compressive light field displays generalize the idea of
content-adaptive parallax barriers to a variety of display architectures, including multiple stacked layers of LCDs (Figures 1
and 5), a thin "sandwich" of two LCDs enclosing a microlens
array or, in general, any combination of stacked, programmable light modulators and refractive optical elements [9]. Similar
to parallax barriers, cascading LCDs usually have a multiplicative effect on the incident light that can selectively attenuate

light in some directions [8]-[12]. The aforementioned outlined
of light field factorization generalizes to all of these display
architectures. Their nonlinear, multiplicative image formation
is fundamentally different from the linear, additive image formation provided by multifocal plane displays, volumetric displays, and many other time-multiplexed displays. In general, a
nonlinear image formation has the potential to provide more
degrees of freedom for the image generation algorithm than an
additive, linear image formation [9], [11], [13].
With a factorization framework for generalized parallax
barriers in hand, applications to a variety of displays other than
television-type systems can be explored. For example, light
field projection systems supporting parallax and stereo cues
have emerged over the last decade [14], [15]. These types of
display systems are most suitable for collaborative experiences and provide impressive image quality over large depths of
field. Unfortunately, dozens of projectors have to be employed,
making multiprojector light field displays expensive, difficult
to calibrate, power hungry, and bulky. The compressive light
field methodology has been shown to also apply to projection
systems [4]. In this case, the goal is to "compress" the number
of required devices, thereby improving power efficiency, form
factor, and cost of the system. Hirsch et al. demonstrated that
this is possible by generating a light field inside a single projection device, via content-adaptive parallax barriers, and then
optically amplifying the limited field of view of the emitted
light field using a screen comprising an array of microscopic
Keplerian telescopes: one in each screen pixel.

IEEE SIgnal ProcESSIng MagazInE

September 2016

125

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