My last post
introduced high dynamic range
concepts and some background to the technology. We discussed the concept of
‘whiter whites’ and ‘darker shadows’
with the ability of modern display technology that can not only output more
light but also increase the dynamic range of the displayed image by reducing
the minimum black level.
SMPTE standardized a HDR EOTF
(electronic to optical transfer function) called PQ (perceptual quantization)
as ST-2084. The PQ transfer function has been optimized to cover a wide range
of light values (from .0001 to 10,000 cd/m2
minimizing the visual effect of 10-bit or 12-bit quantization (contouring).
We touched on an alternate approach
called hybrid log gamma (HLG) transfer curve. HLG is not associated with a
specific light value(s) but rather is a relative light value based on an
assumed dynamic range and peak white. HLG can be used as an image capture curve
as well as the final display transfer curve. While HLG has no metadata
associated with the HDR signal, one has to have an agreed upon peak white
reference value (typically 1000 cd/m2
) for a display to
be able to process the HLG HDR signal and render the image appropriately for
the given display capabilities.
I promised to talk about the
distribution of HDR video over a variety of channels such as Blu-ray, OTT, OTA,
satellite and cable in the this next article. Each mode of distribution has
it’s own unique challenges and options to delivery of video content.
A common element for the delivery of
HDR is HEVC (high-efficiency video coding). The latest video codec from MPEG
not only has the ability to encode 4K video but enables full 10-bit resolution
to the consumer display. HEVC also supports wide color gamut, defined in BT.2020. However, HEVC
still relies on the non-constant luminance equations (YCrCb) that are defined
in BT.1886. HEVC has the ability to signal HDR metadata in a variety of methods
(SEI and VUI metadata) to assist the HDR display to properly process the HDR
imagery. While MPEG-4 AVC does have a mode to support 10-bit encoding, it has
not been deployed in the consumer product space, thus limiting HDR consumer
delivery to 8-bits.
I’ll go over the different
methods of distribution for HDR in the order that they have been adopted and/or
The Blu-ray spec was amended for 4K (UHDTV)
including HDR and wide color (BT 2020). This spec, known as HDR10 is summarized
— HEVC Main level encoding
— (PQ) EOTF (HDR)
— BT.2020 color space
— 4:2:0 subsampling
This format for 4K/HDR has also been
adopted by some over-the-top (OTT) delivery platforms including Netflix and
Amazon. HDR10 is one of the simpler methods of HDR distribution but does
require static metadata (MaxFLL and MaxCALL) to inform the display device of
the average brightness as well as peak brightness values. HDR 10 is not
backward compatible for non-HDR displays, although some Blu-ray players may
provide conversion to SDR if the detected display is not HDR compatible. For
the OTT services, based on the type of display, the appropriate format is
streamed to the display.
LIVE LINEAR BROADCAST
While HDR10 works for optical media and internet
delivery of content, for broadcast channels, HDR10 has some drawbacks with
respect to live broadcasting:
— Requires static HDR meta data
— Requires ‘two
layer’ approach (simulcast of HDR and SDR).
A joint proposal from Samsung, Sharp
and Qualcomm support use of HDR10 for ATSC 3.0. Here are the other proposals being
considered by S34-1 of the ATSC 3.0 Technical Standards Group:
HYBRID LOG GAMMA
(HLG), PROPOSED BY BBC AND NHK
As mentioned above, HLG uses a dual curve
approach, gamma in the dark region and a log function for the bright region. By
optimizing the coefficients of the HLG equation, the tone mapping for HDR and
SDR can be accommodated without metadata or additional processing.
However in practice it has been
shown that the optimization leads to limitations in terms of the overall
dynamic range of the HLG HDR signal to protect the SDR signal. In addition,
while there is no defined meta data, there needs to be an assumed reference
peak white level so that the displayed image tonal range can match the image as
‘graded’ by the video operator. Finally, there is the
challenge of color space conversion between BT.2020 and BT.709
(HDTV color gamut).
HLG is documented in ARIB standard
B67 and will be included in an update to ITU BT.1886.
Dolby’s proposal is based
upon the use of the PQ EOTF curve and optionally a new color space with a new
set of color difference equations called ITP (Intensity, Tritanope, Protanope).
ITP, compared to the established YCrCb color
space, has three components, I
similar to C’r) and
( yellow-blue dimension,
similar to C’b). The underlying color space is based on LMS, which is
based on long-medium-short cone color response of human vision. Dolby
summarizes this format as ITP-PQ. The key benefit of ITP is the property of
isoluminance that minimizes the chroma/luma cross-talk that can happen with the
classic Y’Cr’Cb’ non-constant luminance approach
as well as ‘linearize’ hue versus saturation.
In the encoding process, the ST 2084
PQ-based HDR video (along with static ST 2086 metadata) is converted to ITP-PQ
space and subsampled to 4:2:0. Adaptive reshaping is applied prior to the HEVC
encoder to improve compression efficiency. Specific Dolby metadata is combined
with the converted HDR signal and transmitted as part of the HEVC encoding (SEI
On decoding, the signal is processed
through tonal mapping and then converted to full 4:4:4 color. This
reconstructed 10-bit/4:4:4 HDR/BT.2020 signal is converted from 10-bit to
12-bit and then reverse ITP matrix is applied to output HDR RGB.
Additional metadata (ST 2094) can be
created to provide tonal mapping of full range HDR signals to displays with
constrained HDR performance or to SDR displays, either for professional or
PRIME SINGLE, PROPOSED
Prime Single—now referred to as
“Technicolor HDR Distribution solution”—is a single layer
approach that converts the HDR signal to a SDR signal with dynamic metadata to
provide both tone re-mapping as well as color gamut correction. Prime Signal
supports PQ, HLG, Log or SDR input video signals. At the decoding side, Prime
Signal takes the SDR as decoded and with the tone mapping and CRI color
correction metadata can provide HDR outputs signals (PQ or HLG) as well as a native
SDR (without any further processing or meta data).
This approach provides a backward
compatible SDR output, which is determined by the pre-processing encoding
process. The tone mapping and CRI metadata is used to re-create the HDR signal
from the SDR. The Prime Signal metadata is carried within the HEVC data stream
as SEI messages, with an ability to update on a frame by frame basis.
Ericsson has proposed a
pre-processing approach to a HDR10 HDR signal to mitigate errors caused by
4:2:0 subsampling in the HEVC process. Basically the pre-process calculates the
error due to conversion of RGB to YCrCb, quantization to 10-bit and
downsample to 4:2:0 and then compensates the luma samples to minimize errors on
the decode/reconstruction end.
In addition there is an optimization
of the HEVC QP Chroma offset values to mitigate chroma errors.
No support for conversion to SDR or BT.709 color space.
Pre-processing analysis of the input
HDR signal (HDR 10) to create a set of dynamic range adjustment parameters to
minimize errors in the HEVC (4:2:0 YCrCb) encoding. These parameters are
carried in private SEI messages inside of the HEVC bit stream. Similar to
Ericsson’s proposal, no support for conversion to SDR/BT.709 color
While there are common features
between the HDR proposals, there are different approaches to fitting the full
HDR signal into the limitations of HEVC as well as providing a solution for
multiple display and production formats. Key to evaluating these proposals will
be the head to head evaluation of each proposal scheduled for this June at the
ATSC 3.0 S34-1 committee meeting.