IBC: Philips shows off backward compatible single channel High Dynamic Range

[donaldk]

“No-one was talking about HDR two years ago. At Philips Research we actually build our first HDR display back in 2003, a watercooled monster. This in anticipation of the increasing power and efficiency of LEDs”, Joop Talstra, Standardisation Manager Philips Intellectual Property & Standards, told DCI-Forum during a private demonstration of the Philips Single Channel High Dynamic Range coding and transmission system the company has submitted to MPEG for inclusion in a future revision of the HEVC video compression standard.

At the time Philips was still a large maker and vendor of TV sets and Consumer Electronics, but it sold its remaining shares in the TV Business Unit to TPV Technology of China in 2014.

However the concern continues to perform research in the areas of electronics and imaging, developing Intellectual Property for inclusion in standards, and subsequent licensing. The HDR proposal being a response to MPEG’s February 2015 Call for Evidence (CfE) for HDR and Wide Colour Gamut (WCG) video coding, to explore whether the coding efficiency and/or the functionality of HEVC can be significantly improved for HDR and WCG content.

According to Talstra backward compatibility systems, especially his, perform better than SMPTE 2084, Open HDR or HDR 10, three names for the same Society of Motion Picture Engineers backed standard for HDR that is compulsory in the upcoming 4K Blu-Ray standard, where backward compatible HDR systems that can simultaneously cater to both HDR and SDR displays, like Dolby Vision, Technicolor Prime, and Philips Single Channel HDR, are optional. Blu-Ray has plenty of capacity to offer two separate versions for SDR and HDR so has no direct need for backward compatible HDR, unlike broadcasters. The same goes for first movers in 4K and HDR, streaming video providers.

But according to Talstra these would also benefit from selecting the Philips system. “They also have an interest in things not getting too far out, as too many versions complicates service management, and leads to cache pollution”.

Then Talstra proudly pulls up a slide summarising the results of the testing of five of backwards compatible, and three non-backwards compatible coding systems for HDR over regular (SDR optimized) MPEG HEVC compressed distribution channels. That was presented at the last MPEG meeting in Warsaw in July.

The orange line p22 at the top or near the top among a few others, depending on wich of the three test scenes you look at, was identified as by Talstra as 'that's us'.



Values on or above the horizontal blue dashed line provide statistically significant visual quality superior to the Anchor, while values on or below the horizontal red dashed line provide statistically significant inferior visual quality when compared to the Anchor.

“So perform better than HDR-only and get backward compatibility on top”, Talstra pleads his case.

He continues: “Counterintuitive, right. This is because you work in an SDR world, your MPEG compression chain is fully optimised for Standard Dynamic Range, so systems that make HDR look like SDR will fare better, than pushing pure HDR through an MPEG HEVC encoder”.

How to make HEVC encoders perform better with HDR signal is the subject of several presentations by compression equipment vendors at the annual SMPTE conference in Hollywood late October.

This are tests on how well the various HDR schemes work with MPEG HEVC compression,so not a test of how they grade HDR itself. Of course this is important for real world broadcast and distibution systems as these are all compressed.



"Sequences such as ShowGirl2 and Market3 are good for testing HDR compression. On the other hand, sequences with a wide dynamic range and strong luminance temporal changes, such as AutoWelding although good for demonstrating HDR may not be necessarily best to assess HDR compression performance".

All the proponents more or less performed on an equal footing to HDR 10.

The MPEG testreport continues: "Dark scenes are important too, as HDR is not only about high brightness, but it might be hard to see the improvements in these sequences, especially if the previous test sequence was bright, due to the adaptation time of the human eye".

The full report on the paneltests results on how the five backward and three non-backward-compatible HDR only proposals stack up against HDR 10 by MPEG may be found here: MPEG HDR HEVC compression test results.

The associated Electro-Optical Transfer Function, for HDR, which tells how to turn digital code words into visible light, a replacement for traditional Gamma, that covers both darker and much brighter images, is part of an ongoing standardization effort at the ITU, that considers EOTF enhancements to the standard Gamma of ITU-R 1886 corrently used in the UHD standard ITU-R. 2020. This workgroup is where BBC R&D, Dolby, Technicolor  and Philips have introduced their transfer curves. Standardisation at the ITU offers vendors  another opportunity following SMPTE and the Blu-Ray Association for standardization and IP monetisation.

Philips has also developed a coding scheme, Y”u”v”, that seperates light and color, that should help with High Dynamic Range imaging as the brightness no longer depends on color value. So colors can be coded as being both bright and saturated at the same time. This goes counter current practice of coding light as color-difference signals. Introduced at the IBC 2014 Conference, this has yet to gain much traction among the broadcast and content industry. According to Talstra this “has always been a long term proposition”.

Philips' HDR coding system is currently ready for implementation and licensing, says Talstra. Asked if this could be done in software or would require a hardware redesign, Frédèric Guillanneuf, Manager Business Development for Philips Intellectual Property & Standards, said: “this is possible on some chips, but would be very expensive, it would cost too much power, so it really is for hardware implementation”. In hardware it requires only a (very) small portion of the video decompression/processing chip. Hardware implementation should be straightforward, as Talstra explains: “what is generally difficult to implement (in chip hardware) are large area blocks, and high I/O”, this is neither. Chip design involves high upfront cost. Talstra confirms this is therefore ‘an high volume proposition’.

And that is what is currently at stake, as a handful of proponents of HDR coding and transmission systems are currently vying for adoption of their system at various arena’s of standardisation, MPEG, ITU, DVB, and of course UHD Blu-Ray. This makes that CE vendors and by that chip vendors are hard to commit. For instance Samsung announced its support in a press release for the BBC R&D Hybrid Log-Gamma technology, saying it would continue to develop HDR support on the basis of this technology. However at IBC all the Samsung SUHD displays used just the HDR10 scheme. LGs OLED as demoed at the SES booth at IBC, I am told, even ran on the BBC technology, but LG changed that for the demo at IFA and IBC. And this is just the no or low cost BBC(-NHK) technology, not the commercially licensed systems from Dolby, Technicolor or Philips. Imagine supporting five HDR systems…

[donaldk]

So how does the Philips Dynamic Range Conversion work. It all starts with an HDR source, a color grader creates a SDR version from this HDR graded material. “This ‘artistic info’, which knobs you turn to convert HDR to SDR, is what you record, that’s the (dynamic) metadata”, Talstra explains the system. Philips proposal is all about preserving artistic intent and artistic control. Though it allows for presets and limited adjustment real-time, live workflows, such as live sports coverage.

"In a real time workflow, as used in live broadcast, typically only one (shadow gain control) or maybe two (plus highlight gain control) parameters would be adjusted on the fly. For all other parameters, default values would be used", according to a Philips whitepaper on its HDR system.

The 'dynamic metadata' describing the difference information amounts to less than 10 kilobit/s per second. It is transmitted with the grader created SDR video signal where a set-top or display hosting the Philips dynamic expansion algorithms on a larger processing chip re-generates the HDR version. The display will most likely vary in its peak brightness and colorvolume, and other capabilities, from the master display, so static metadata portraying its capabilities to the HDR renderer that adjusts the HDR output to the TVs HDR capabilities.

"The dynamic range conversion at the display side takes the transmitted SDR signal and this metadata to render the HDR signal. This describes how to go from the 100 nits SDR version to the level of the HDR master display, in this case a 4000 nits Sim2 LCD. He continues: “metadata tells the direction, tells you to cross over, info from the TV tells you the length, or how far to cross over to the 4000 nits reference”, as Talstra draws a line from 100 nits SDR to the 4000 nits master display, with a 700 nits TV sitting somewhere on the line. "of course this is not a single direction, nor a linear relationship. This negotiation takes place in Philips algorithms currently on an FPGA, but eventually a video processing chip in a set-top-box or display.



The broadcasted signal may also be the HDR grade, with the dynamic metadata describing how to generate the SDR version. This requires all receivers to contain a HDR 'decoder or renderer', is less popular with broadcasters, that prefer a fully backward compatible system using 'legacy infrastructure'. This mode results in optimal HDR, with all the grading compromises such as clipping whites shifted to the SDR grade. So from a technical perspective this is considered better, but demand is for backward compatibility, so Philips tools offer options to avoid errors when creating the SDR grade, retaining enough information to re-generate the HDR version.

When asked about automated conversion Talstra promptly responds "Your Milage will vary". It all depends on the quality and of your source material.This is even more the case with archival material, that has been graded for SDR. "Again your milage will vary, sometimes it will come out okay, other times it becomes extrememly labour intensive to correct the results", Talstra ads speaking of a few tests done with 'studio's'.

Philips recently published a good whitepaper on its HDR grading and coding system.

A summary from this whitepaper:

The process starts from an HDR video master. This HDR master must be graded on a high-dynamic-range
display with high peak luminance and low black level. In the case of live production, such display should
be used to monitor the HDR video master. The HDR video master is encoded. Typically, an HEVC video
encoder will be used, with SMPTE ST 2084 as EOTF.

An SDR video master is derived in a semi-automatic way from the HDR master. First, an initial tone
mapping is proposed. This may be based upon an average tone mapping curve, or may be created by an
automatic analysis of the input image. A colorist then adds corrections to this initial to optimize the SDR
master. This is typically done on a scene-by-scene basis, but may be done on a frame-by-frame basis if
desired. This ‘artistic guidance’ process is typically done in a trim-pass.

The corrections by the colorist are captured in conversion parameters, which are transmitted as
content-dependent metadata along with the encoded HDR video using SEI messages embedded in the
video stream. The content-dependent metadata is defined in SMPTE standard ST 2094-20. The
characteristics of the display used for grading or monitoring, such as peak luminance and black level, are
added as SMPTE ST 2086 metadata to the video stream.

At the receiving side, the video stream is decoded, producing HDR video. The conversion parameters are
extracted from the SEI messages. If an SDR output signal is desired, exactly the same dynamic range
conversion process as set by the colorist on the encoder side is repeated in the decoder, producing an
SDR signal that is - apart from artifacts introduced by the video encoding-decoding - equal to the SDR
master on the encoder side. If an HDR display is connected, the conversion parameters are used to
produce an HDR output signal optimized for the specific display capabilities of the display.<\q>

[donaldk]

The demonstration configuration was fairly straightforward. A laptop plays out 1080P material over HDMI, an convertor converts this signal to SDI to feed an MPEG HEVC Main10 4:20 encoder, in this case an Keepixo.





The SDR graded HEVC compressed video signal including around ten kilobit/s of metadata embedded in the transport stream are fed to an ViXS 4K and HDR capable set-top box, that sends the signal directly to the Philips SDR LCD TV and to an FPGA board that takes in the static metadata from the display and the dynamic metadata from the Philips system to determine the HDR output. The board plays out HDR at the reference level, 4000 nits, to the Sim2 master(ing) monitor. And at an adjusted based on the static metadata describing the display capabilities from the 700 nits Samsung SU 9500 HDR TV.

The SDR TV with its feed graded for peak whites at 100 nits (Candelas per square meter), is driven so hard to reach around 350 nits that it begins to loose color in the bright area's, the orange red flames on the left display are whited out and show no blue the color of the hotter flames. Also there is less texture in the flames, so flames are less discernable.


The Sim2 4000 nits peak brightness HD monitor shows even more color and more texture and detail in highlighted areas.



Current state of the art 700 nits HDR TV from Samsung, showing HDR content adjusted to the TVs capabilities in brightness and colorvolume, by the Philips algorithm running on the Xilinx FPGA development board,  versus the full 4000 nits HDR graded materialon the Sim2 monitor.




Generally the Sim2 shows more color, and therefor more detail withing the bright colored areas. Slightly more blue flames underneath and among the orange red ones. Though that seems less clear in the photographs than in real live.






The SDR TV shows less defined wrinkles in the water, the HDR TV shows more detail and highlights in the branch reflected in the water. And when witnessing the demo live, the reflection of the sun goes from white blob to more and more a sun that is also a bit richer yellow instead of the white.on the regular TV.





The concrete of the stadium wall clearly shows more color information, showing more depth on the HDR monitors than the SDR TV. The sun behind the cloud is one big blob of white light on TV, clouds and a sun start to appear on the Samsung HDR TV and on the Sim2 4000 nits HDR monitor which shows the sun behind the cloud.





White on white in bright areas also show clearly in a scene of a white haired gentleman sitting at a white clothed restaurant table in front of cream white curtains, in front of a window letting in bright sunlight. The greater the dynamic range the greater the grading in texture, and how creamy the white of the curtains is becomes clear with the full dynamic range upto 4000 nits.