IEEE Consumer Electronics Magazine - October 2015 - 79

Camera
Input Light

Camera
Sensor

Data
Reconstruction

Raw Data

Color
Pixel Encoding

Camera Output

Standard Color Pixel Representation
BT.709/ DCI P3/BT.2020
BT.1361 (Gamma) /SMPTE ST 2084 (PQ)

Open EXR/RGBE/XYZE

FIGURE 12. An acquisition pipeline from raw camera data captured by a camera sensor to a standardized representation.

doing this is by superimposing a spatially varying neutral
density filter on top of the camera sensor [26]. Spatial bracketing techniques usually trade off resolution for higher
dynamic range. Note that modifying a sensor (like superimposing a neutral filter) is a costly operation.
In multisensor bracketing, multiple sensors are embedded
in a single camera along with a beam-splitter to divide the light
between the sensors and achieve different exposures [27]. This
method requires changing the optical system, which is usually
the most expensive part of the recording device. Furthermore,
as several sensors need to be embedded in the camera, space
limitation and portability of the camera might be an issue.
Multicamera bracketing involves several cameras (usually
two) with a null baseline rig. An appropriate neutral gray filter is put in front of one of the cameras to generate two views
with different exposures. Before combining the exposures for
creating HDR content, the geometric disparities (due to sensor, mirror imperfections, and optics differences) need to be
corrected. This technique provides good results, but the portability of the camera setup can limit its applications.
All of the above-mentioned HDR acquisition techniques
have their own pros and cons, as summarized in Table 4.
While capturing real scenes in HDR is quite demanding, it is
easy to produce computer-generated HDR video content. As
camera sensor technology improves, single sensors capable
of capturing the wide range of visible light and color gamut
are likely to become available in the near future, which eliminates the need for bracketing techniques.

the redundant information in the video signal and only
keep the information that is essential to reconstruct the
signal on the decoder side.
To compress HDR video content stored in floating point
values, three main approaches have been proposed. The first
main category is scalable and requires two codec instances:
one to compress a tone-mapped (LDR) version of the original
HDR content and a second one to compress the missing
information between the tone-mapped (LDR) version and the
original HDR one. The missing information can be, for
example, a quantized ratio between the original HDR and its
associated tone-mapped LDR version [28] or the residuals
between the original HDR and an inverse tone-mapped version of the associated LDR version [29]. At the decoder side,
in the former case, the decoded LDR content is multiplied by
the decoded ratio content, while in the latter, the HDR version is reconstructed by adding the residue to the decoded
inverse tone-mapped content. The main advantage of these
techniques is the ability to provide two types of content: an
HDR and an LDR one, thus supporting both HDR displays
and legacy LDR displays. However, their main drawback is
the use of two codec instances, with one (the HDR codec)
that may require more than 8 b for a good HDR reconstruction. These techniques correspond to the scalability scenario
as described in Figure 18(b).
The second group of HDR video coding techniques is
based on using a perceptual curve (for example, the PQ [18])
that transforms HDR content (floating-point data) into LDR
content of high bit depth (integer data with quantization step

VIDEO COMPRESSION
5,000
Encoded 12-B Value

Color pixel encodings reduce the amount of information by
optimizing the use of integer code values and by restricting
the range of light and color that can be represented for
storage or distribution purposes. Tables 5 and 6 summarize
the bit rates required to store uncompressed videos in different encoding and the targeted bit rate for digital broadcast as well as the storage capacity of a legacy storage
disk. The ratio between the broadcaster's bandwidth and
the uncompressed data bit rate is of the order of 150
(1.25 Gbps/8 Mbps). To reduce the size of video data and
meet the bit-rate requirements of broadcasting or disk
storage-capacity limitations, video compression techniques are used. Video compression techniques remove

4,000
3,000
2,000
1,000
0
10-4

10-2
100
102
Luminance in Log-Domain (cd/m2)

104

FIGURE 13. Luminance values encoded with the PQ [18] using 12 b.

october 2015

IEEE Consumer Electronics Magazine

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - October 2015