An Introduction to Hybrid Log-Gamma HDR Part 1: Mixed Display and Viewing Environments

Presenter: Andrew Cotton (BBC)
Duration: 12 min
Slides: download

All talks

Slides & video

Hello, I'm Andrew Cotton.

I'm a Principal Technologist at BBC Research and Development.

I'm also one of the developers of the Hybrid-Log Gamma HDL solution.

And that's what I'm gonna be talking about today.

This is part one of my two talks.

And in this part, I'm gonna be providing an introduction to HLG explaining the design principles.

And also I'm going to explain how it was designed from the outset so I'll get a wide range of different capability HDR displays watched across a wide range of different viewing environments.

In the second talk I'm gonna focus on format conversion and image compositing.

So Hybrid-Log Gamma gets its name from the camera OETF.

That's the opto electrical transfer function.

And that's illustrated on the graph on the slide.

Now, the job of the OETF is to take the linear scene lights signal produced by the camera sensor and convert that into a nonlinear signal suitable for 10 or 12 bit digital interfaces.

And the curve itself is carefully matched to the sensitivity of the human visual system to minimize banding artifacts after that quantization.

In the case of HLG, the OETF comprises a traditional gamma curve in the lower part of the signal range.

That's very similar to the curve found in SDL cameras and there's a logarithmic curve spliced onto the upper range of the transfer function to extend its capability in the highlights.

Now, because the lower part of the curve is the same as a conventional SDR camera curve, and that's the part of the signal range which is most important for our subjective impression of a scene, we're able to take an HLG signal and present it on a conventional wide color gamut SDR display and get a good quality backwards compatible picture.

But again, I'll talk more about that in part two of the talk.

So having focused on the O E T F it's actually the end to end opto optical transfer function.

That's the most important part of any TV system and the OOTF comprises the camera OETF and the almost opposite function, the display EOTF, concatenated together.

Now important to note, that the OETF and the EOTF are not exact inverses of one another.

And that's because the job of the OOTF is actually to ensure that the images seen on the display are subjectively similar to the scene in front of the camera And what that means is that the OOTF actually needs to vary for different brightness displays and different display surrounds because both of those aspects affect the adaptation state of the eye, which changes the way that we perceive tones.

Now, the OOTF is usually a gamma law, and you might wonder how does a gamma law actually ensure the subjective appearance of pictures on the screen?

I certainly did.

And by a gamma law, we mean that the displayed light is proportional to the scene light raised to a power gamma.

And the reason that the gamma law works so well is because the eyes sensitivity to light is approximately logarithmic.

So if we then take logarithms of both sides of that equation to convert it into units which sort of approximate what's interpreted by the brain, you can see that the log of the display light then becomes directly proportional to the log of the scene light.

And that's plotted on the graph on the right.

And because they're linearly proportional to one another you can see that this gamma law is able to maintain the subjective appearance of tones across a wide range of different brightness displays.

That's important to get the value of that OOTF gamma correct, if it's too low, the pictures look washed out.

As you can see on the image on the left.

If it's too high they look a little bit overly colorful and a little bit too punchy.

It's been known since I think the 1940s that brighter viewing environments require a lower OOTF gamma.

And if you think about it, that's quite obvious because the lower gamma will mean that detail in the shadows of an image becomes brighter, and that's actually what you need to do, If that that detail is to be visible as the brightness of the surround increases.

But it actually wasn't until very recently when we conducted a number of subjective tests at BBC R&D that we knew exactly how that gamma needed to vary for different display surrounds.

And the, those experiments are described in ITU Report, BT.2390 and the formula for changing the OOTF gamma based on the luminance of the display surround is seen on the bottom part of this screen and described in BT.2390.

And that's actually, that gamma adjustment is now implemented in some consumer TVs.

Whilst, many of us knew that the gamma needed to be lower for brighter viewing environments, what we didn't know was how that gamma needed to change for different brightness displays.

Indeed, people used to think the brighter the display, the closer to nature it is, and the lower the value of gamma that you needed.

But we actually discovered quite by accident, it was the other way around.

We were giving a demonstration at IRT in Munich in June 2014, and we had a bright digital signage display that we had carefully calibrated and and used a 3D lookup table in front of it to convert that those, that that system into an HLG display with a fixed camera of 1.2 which is what we had for the reference viewing environments at the time.

And the pictures just looked wrong.

They were washed out and overly bright.

So when we got back to base we rapidly did some rough and ready tests and found that actually the gamma needed to get higher for brighter displays.

And then we did two further sets of subjective tests, and NHK did a set of subjective tests.

And all three of those formal tests confirmed the relationship which is illustrated in the graph on the bottom right of this slide and reproduce both in ITU report 2390, but also in the normative part of the recommendation BT.2100, which describes HLG.

In fact, there are two forms of that equation.

There's this form shown on screen in note five F but actually there's a more accurate formula in footnote two of BT 2100 that works better over a wider range of display peak luminances.

So that's the one that I'd recommend using from now on.

And actually just very quickly, if we plot the log of the scene light across the log of against the log of the display light for a range of different peak luminescence displays and we want to maintain this linear proportionality between those two logarithmic quantities, you can very quickly see that actually for brighter peak luminance displays you need a higher value of gamma.

So actually this should have been obvious right from the outset.

It just wasn't, it wasn't to us anyhow.

So what does that mean?

For scene referred systems like HLG actually you can consider that the OOTF is implemented solely in the display and that's because if we drive down into the end to end system you can see we've got the camera, which takes scene light, puts it through an HLG OETF to give us a nonlinear signal suitable for transmission over 10 or 12 bit interfaces within the display itself.

The first thing we do actually is undo that nonlinearity which was simply there to make the use, make best use of a 10 or 12 bit signal chain.

And that then gives us the scene light signal again.

And it's that scene light signal within the display itself to which we apply an OOTF.

For PQ, because the nonlinear signal represents the brightness of the pixel on the display, it's actually the other way around.

And if you do the same sort of analysis, what you find with PQ is that the OOTF is effectively implemented within the camera or the grading suite.

So the OOTF for that reference grading environment or production environment is effectively baked into the signal.

And that's the two differences.

And that's the key differences anyhow between the two systems.

So we take those two aspects together, with HLG we've got one signal and simply by changing the OOTF gamma implemented within the display we can present that signal on a range of different brightness displays, across a range of different viewing environments.

So if we take the value of 1.2 which is highlighted in the table in red, that's for a reference display in a reference viewing environment with a five candela per meter squared surround, we can take that same signal and put it on a laser projector at 108 candelas per meter squared in a really dark viewing environment with the hats at 0.05 candela per meter squared surround with a gamma of 1.1 and we can get remarkably good pictures that look just like those on the TV type monitor in the reference TV viewing environment.

And we've done that sort of demo many times.

Er, several times, I should say, and some of you may have seen it at IBC.

If you want to know more about different viewing environments than do read the BBC blog that's linked to on the bottom of this slide.

Before I wrap up, I just wanted a word about HLG in different application areas.

I'm sure most of us know that the big streaming platforms such as Netflix and Amazon Prime use PQ for HDR systems.

What you might not be aware of is that amongst TV broadcasters, it's actually the other way around.

The vast majority of broadcasters have adopted HLG as have the cable IPT operators as well.

So in, in these figures here.

And the other thing that might come as a surprise was back in October, I think, of last year we were delighted to see the iPhone 12 was launching with HDR, video recording, using Dolby Vision.

What you might not have known is that particular variant of Dolby Vision, profile 8.4, was built on, is built on HLG not PQ as the nonlinear transfer function.

So that's it from me, just to summarize HLG from the outset was designed to cater for a wide range of different display, er, display luminances and a wide range of different viewing environments.

There are two OOTF gamma adjustments specified by the ITU.

One for the peak luminance of the display, the other for different surrounds.

And actually HLG is becoming the HDR format of choice amongst TV broadcasters.

Thanks so much.

All talks