or Connect
AVS › AVS Forum › Display Devices › Display Calibration › How power law gamma calibration can lead to crushed blacks
New Posts  All Forums:Forum Nav:

How power law gamma calibration can lead to crushed blacks - Page 2

Quote:
Originally Posted by wmwilker

You mentioned "what the colorist/director sees". Is that based on a flat gamma?

That's exactly what we are discussing, whether or not what we watch was mastered using a pure power law gamma system. I don't know that anyone can answer that definitively.

Edited by zoyd - 12/2/13 at 2:16pm

AVS Top Picks

Quote:
Originally Posted by zoyd

That's exactly what we are discussing, whether or not what we watch was mastered using a flat gamma system. I don't know that anyone can answer that definitively.

Sorry, I thought that after I'd asked.
If you'd like to try to emulate a raw Sony BVM CRT use the following target luminances for your gamma adjustment points. They are scaled to 120 cd/m^2 peak white so multiply all values by your_peak_white/120 to scale. First column is calculated from the calibration data provided for the monitors used in the Filmlight system, second column is an average of the Filmlight monitor and a flat 2.4 monitor.

 %Stimulus BVM BVM-2.4 Average 5 0.11 0.10 10 0.40 0.44 15 1.08 1.17 20 2.33 2.42 25 4.31 4.31 30 7.14 6.90 35 10.87 10.25 40 15.50 14.36 45 21.02 19.27 50 27.37 24.95 55 34.48 31.39 60 42.27 38.58 65 50.67 46.50 70 59.58 55.11 75 68.94 64.40 80 78.66 74.33 85 88.69 84.88 90 98.96 96.03 85 109.41 107.74
Quote:
Originally Posted by Chronoptimist

The target for an ideal display is a "flat" 2.40 gamma. Sony's OLED monitors are capable of this, and that's how they are calibrated. Sony's CRTs were not. (0.01cd/m2 black level)

If your display has less than perfect black levels, the BT.1886 transfer function compensates for this.

If you want to emulate a CRT (and there's no real reason for an end user to do so) there is the alternate EOTF as defined in BT.1886. Most CRTs will be LUTed to as close to 2.40 as possible these days.

That is assuming the mastering studio is using those monitors. When I talked to the master people at Sony, they are mastering to a 2.2 spec. I've heard some studios are using 2.4. You also have to take into account that gamma is dependent on how bright you are at 100 IRE, which you don't know what their reference was, and you have to take into account ambient, which you may not know.

Gamma sucks, always has always will. Nothing more frustrating if you're trying to be "absolutely accurate".
Quote:
Originally Posted by Kris Deering

Gamma sucks, always has always will. Nothing more frustrating if you're trying to be "absolutely accurate".

The trick is that unless you're including ambient light, display max and display min in you gamma formula you really don't have a chance or matching perceptually.

Granted if those variables aren't part of the formula, you can simply replicate them.
Quote:
Originally Posted by zoyd

Regarding analog film to digital transfers, are you saying the DCI specs are followed as normal practice in the transfer to DVD/BD or were you referring to digital theatre projection?

Theatre projection. Transfers to DVD and BD are very much video world rather than DCI.

For some more industry input I recommend asking the people on either the cml-post or TKcolorist mailing lists, both of which have been host to discussions about gamma and reference monitoring over the years.

At cinematography.net, join the cml-post list
At tig.colorist.org, join the tig mailing list

I think those are the places where you'll find most colourists and chief engineer types on the net.

It's hard to get people to stick their necks out and say "this is what we should be calibrating to", but it might be possible to find out what displays people are preferring to replace the Sony BVMs.

Manufacturers have a lot of sway here too - it would be interesting to see what transfer functions are used out of the box on the newer Sony, Dolby, JVC, Barco, Panasonic, Cinetal et al reference monitors - the problem is that there are so many to choose from now!

Best,
L
Thanks so much for your input! Perhaps it will become a non-issue if BT.1886 gets widely adopted but probably not given the legacy of not defining a standard up-front.
Quote:
Originally Posted by Kris Deering

That is assuming the mastering studio is using those monitors. When I talked to the master people at Sony, they are mastering to a 2.2 spec. I've heard some studios are using 2.4. You also have to take into account that gamma is dependent on how bright you are at 100 IRE, which you don't know what their reference was, and you have to take into account ambient, which you may not know.

Sony's LCD monitors were 2.2 out of the box, simply due to the fact that they don't have the contrast ratio required for anything more. In fact, if you use the standard BT.1886 with an 0.1cd/m² Lb value, it comes out very close to 2.2 gamma.

Standard monitor brightness is 100cd/m² almost universally now. (traditionally it has ranged from around 80–120cd/m²)

And IRE is no longer valid, that's an old analogue term.

Quote:
Originally Posted by lewis.saunders

Manufacturers have a lot of sway here too - it would be interesting to see what transfer functions are used out of the box on the newer Sony, Dolby, JVC, Barco, Panasonic, Cinetal et al reference monitors - the problem is that there are so many to choose from now!

Well I can tell you that Sony's OLED monitors are using a flat 2.40 out of the box, and Barco's monitors use a flat 2.35 out of the box. (based on the older EBU recommendation)

Quote:
Originally Posted by zoyd

Thanks so much for your input! Perhaps it will become a non-issue if BT.1886 gets widely adopted but probably not given the legacy of not defining a standard up-front.

Things are moving in that direction. However, the fact remains that regardless of how a film was mastered, for the best viewing experience at home, you should follow the standard BT.1886 formula, as it takes your display's capabilities into account. Running a flat 2.40 gamma on most flat panels is a good way to ruin the image for example, as they just don't have the contrast ratio required.

Until there is a widely adopted standard, you could debate this point endlessly. The fact is that very few displays out there are capable of emulating a broadcast CRT—they just don't have the contrast ratio for it, to start with. Compensating for that using the alternate BT.1886 function may match the appearance of a CRT with its brightness raised to that level, but the fact is that those CRTs are not used in that way.

As we move forward, and more CRTs are replaced, there is going to be far less of a need to try and emulate them. Even with gamma curves that emulate a CRT response, the fact is that a CRT is a more "dynamic" display, and the response changes with the image being displayed on it, to some degree.
Just to have a look, here are the two BT.1886 transfer functions calculated for my display (peak white 100 cd/m^2, mll 0.05 cd/m^2). It's interesting that in order to reach a gamma=2.35 over the flat portion of the primary recommendation curve requires an mll of 0.005 cd/m^2

when is the new/next standard for gamma going to come out and what will it look like versus the current power formula? what will it be called?
The recommendation has been published, I don't know how such things are adopted in practice. The plot above shows you what the gamma profile will look like based on the recommended transfer functions (there are actually two recommendations, an alternate one if you desire to better emulate a "typical" CRT response). The magnitude and to some extent the shape of the curve depend on the display setpoints - mll and peak white.
Quote:
Originally Posted by sotti

The only thing that matters is that transfer function on your displays is a perceptive match to the the function used on the display where they graded the master.

Theory is fun to play with, but the rubber meets the road in the actual studios. You want to do what they are doing. Eventually they'll likely all be using BT.1886, right now they are using power typically with an exponent of 2.35.

will CalMANv5 support the new formula? is there a reason 2.35 is currently being used for the power formula versus 2.2 or 2.4 or something else from 2.2-2.6?
Quote:
Originally Posted by zoyd

Just to have a look, here are the two BT.1886 transfer functions calculated for my display (peak white 120 cd/m^2, mll 0.05 cd/m^2). It's interesting that in order to reach a gamma=2.35 over the flat portion of the primary recommendation curve requires an mll of 0.005 cd/m^2

http://i64.photobucket.com/albums/h1...1at12143PM.png

Those are... interesting results for the alternate CRT-matching function. Are you sure they are correct?

I've also been plotting various transfer functions for comparison, but have been having trouble with that one. If I'm reading things correctly, b should be manually adjusted until L for V=0.0183 is equal to measurements taken from the reference display? What data did you use for the reference display?

Does the alternate CRT matching function work when Lw is different? (e.g. when you have your screen brightness at 120cd/m² rather than the 100cd/m² reference level)

I'm also confused because the paragraph preceding the alternate transfer function makes mention of Lb, but it is not actually referenced in the function itself. (it is in the reference electro-optical transfer function, however)
Quote:
Originally Posted by PlasmaPZ80U

will CalMANv5 support the new formula? is there a reason 2.35 is currently being used for the power formula versus 2.2 or 2.4 or something else from 2.2-2.6?

Yes BT.1886 is in CalMAN 5.
Quote:
Originally Posted by Chronoptimist

I'm also confused because the paragraph preceding the alternate transfer function makes mention of Lb, but it is not actually referenced in the function itself. (it is in the reference electro-optical transfer function, however)

I had similar issues in reading the alt-EOTF function. Since effectively no one is using BT.1886 yet, I can't imagine the alt function is going to be used anywhere in the real world.

Due to very similiar concerns CalMAN 5 won't have Alt-1886 turned on at least initially.
Quote:
Originally Posted by Chronoptimist

Those are... interesting results for the alternate CRT-matching function. Are you sure they are correct?

I've also been plotting various transfer functions for comparison, but have been having trouble with that one. If I'm reading things correctly, b should be manually adjusted until L for V=0.0183 is equal to measurements taken from the reference display? What data did you use for the reference display?

Does the alternate CRT matching function work when Lw is different? (e.g. when you have your screen brightness at 120cd/m² rather than the 100cd/m² reference level)

I'm also confused because the paragraph preceding the alternate transfer function makes mention of Lb, but it is not actually referenced in the function itself. (it is in the reference electro-optical transfer function, however)

I checked it and did have an error in b, it was too small. I reset it such that L=mll (0.05 cd/m^2) at v=0.018 since that level is below my mll on a 2.2 display anyway. After I did that I got funky results at the high end so reset peak white to 100 cd/m^2 and replotted above. Shape is about the same.
Quote:
Originally Posted by sotti

I had similar issues in reading the alt-EOTF function. Since effectively no one is using BT.1886 yet, I can't imagine the alt function is going to be used anywhere in the real world.

It seems to me like the Reference 1886 transfer function is what is going to be adopted going forwards. The "alt-1886" function seems to be for edge-cases where someone needs to match a specific monitor in their grading suite for a very specific reason. (perhaps they were in the middle of a project and needed to get a new monitor due to hardware troubles?)

While I understand that someone may wish to try and emulate a CRT, it seems unlikely to happen moving forward, and is more for legacy support while the industry is making a transition to newer monitors.

Especially for end-users in the home, the reference transfer function seems ideal, when it also compensates for display contrast capabilities. I think it's a far better solution than we've ever had, where it's often been up to the calibrator's preferences when dealing with ambient light, and so on.
Quote:
Originally Posted by Chronoptimist

Especially for end-users in the home, the reference transfer function seems ideal, when it also compensates for display contrast capabilities. I think it's a far better solution than we've ever had, where it's often been up to the calibrator's preferences when dealing with ambient light, and so on.

I don't agree. It bothers me that the recommendation is based on display contrast ratio and ignores rendering intent. My display can hold a 2.4 gamma down to 4% stimulus (actually closer to 3% for ansi contrast) and yet this formula tells me to target a shaped curve from 2.1 to 2.25. If something that was originally mastered at 2.4 is displayed on this display, with that calibration, the colors will be shifted.
Quote:
Originally Posted by zoyd

If something that was originally mastered at 2.4 is displayed on this display, with that calibration, the colors will be shifted.

But will your perception of it be altered?

The big issue here is that the x,y,Y (XYZ) colorspace doesn't tell you anything about how saturated you will perceive the color. The perception of color is relative to the viewing environment. Even L*a*b* is only so good about compensating for environment.

If anything BT.1886 doesn't go far enough in environmental compensation.
Quote:
Originally Posted by sotti

But will your perception of it be altered?

The big issue here is that the x,y colorspace doesn't tell you anything about how saturated you will perceive the color. The perception of color is relative to the viewing environment.

If anything BT.1886 doesn't go far enough in environmental compensation.

Agreed, I was just baselining the problem in matched environments. There's no reason to think it would get any better once you factor in surround.
Quote:
Originally Posted by zoyd

I don't agree. It bothers me that the recommendation is based on display contrast ratio and ignores rendering intent. My display can hold a 2.4 gamma down to 4% stimulus (actually closer to 3% for ansi contrast) and yet this formula tells me to target a shaped curve from 2.1 to 2.25. If something that was originally mastered at 2.4 is displayed on this display, with that calibration, the colors will be shifted.

Most displays should be capable of a relatively flat 2.4 gamma down to 5% or so. The problem is that when you do this, your only option is to clip all levels below that, which is unacceptable.

Your only options then, are to use a non-linear gamma that ramps up considerably near black, to use a lower gamma, or to use a compensated gamma like the BT.1886 transfer function.

While it may be less accurate, perceptually, the compensated gamma tends to look better.

If you want to be accurate to the mastering process, you need a display of equal or greater quality to the monitor being used, which doesn't yet exist for the home market.

My local-dimming full array LED backlit LCD does a pretty good job with a flat 2.4 gamma. Unlike many displays, the dimming algorithm is not used to boost contrast as high as possible, but rather it is designed to extend the dynamic range of the display. A 10%stim box will remain the same brightness regardless of APL, for example, unlike other displays where the brightness of that box would change with APL to maximise image contrast at the cost of accuracy. (even some Plasmas suffer from that sort of thing) It isn't perfect as blooming comes into play at times... but the same thing happened with CRTs.
Quote:
Originally Posted by Chronoptimist

Your only options then, are to use a non-linear gamma that ramps up considerably near black, to use a lower gamma, or to use a compensated gamma like the BT.1886 transfer function.

While it may be less accurate, perceptually, the compensated gamma tends to look better.

I fully understand that argument but I'm not convinced that it's a better trade. Is a perceptual error of 3-5 dE94 in flesh tones better to have than a too low gamma from 0-4%?
Quote:
Originally Posted by zoyd

I fully understand that argument but I'm not convinced that it's a better trade. Is a perceptual error of 3-5 dE94 in flesh tones better to have than a too low gamma from 0-4%?

I believe it absolutely is.

Crushed black are one of the most obvious errors, it's something even people are even causally acquainted with calibration readily notice.
The degree of compression would have to be evaluated on the particular display against the perceptual error in reference material, as anyone even casually acquainted with calibration would admit.

I recalculated the predicted baseline saturation shifts between a flat 2.4 master and the recommended EOTF for my display and they are not as bad as the raw BVM mismatch.(Assuming matched environments of course!)

 Color dE94 dark skin 2.83 light skin 1.62 blue sky 2.28 foliage 2.56 blue flower 2.05 bluish green 1.30 orange 1.60 purplish blue 2.78 moderate red 1.91 purple 3.15 yellow green 1.32 orange yellow 1.36

Now I think it's beertime again!
Quote:
Originally Posted by zoyd

Now I think it's beertime again!

Couple more hours for us left coasters.
Quote:
Originally Posted by zoyd

If you'd like to try to emulate a raw Sony BVM CRT use the following target luminances for your gamma adjustment points. They are scaled to 120 cd/m^2 peak white so multiply all values by your_peak_white/120 to scale. First column is calculated from the calibration data provided for the monitors used in the Filmlight system, second column is an average of the Filmlight monitor and a flat 2.4 monitor.

 %Stimulus BVM BVM-2.4 Average 5 0.11 0.10 10 0.40 0.44 15 1.08 1.17 20 2.33 2.42 25 4.31 4.31 30 7.14 6.90 35 10.87 10.25 40 15.50 14.36 45 21.02 19.27 50 27.37 24.95 55 34.48 31.39 60 42.27 38.58 65 50.67 46.50 70 59.58 55.11 75 68.94 64.40 80 78.66 74.33 85 88.69 84.88 90 98.96 96.03 85 109.41 107.74

For the math challenged among us, are the numbers in this table accurate? I notice that the 25% stimulus values are equal at 4.31.

Maybe if I understood the formula for calculation of the target Y values, I could create a spreadsheet to generate them. I guess that's what you did.

Also, I have a 9 year old Sony XBR HD (CRT) that I have not done any calibration of since it has little in the way of adjustments and is in the guest bedroom. But I could set it up with a pluge and get a set of calibration measurements to see how it measures up. See what the gamma curve looks like and use the Y values to compare to your suggested (calculated) empirical Y values above.

It's not a BVM of course but it is a Sony CRT ...
Quote:
 Originally Posted by kjgarrison For the math challenged among us, are the numbers in this table accurate? I notice that the 25% stimulus values are equal at 4.31. Maybe if I understood the formula for calculation of the target Y values, I could create a spreadsheet to generate them. I guess that's what you did. Also, I have a 9 year old Sony XBR HD (CRT) that I have not done any calibration of since it has little in the way of adjustments and is in the guest bedroom. But I could set it up with a pluge and get a set of calibration measurements to see how it measures up. See what the gamma curve looks like and use the Y values to compare to your suggested (calculated) empirical Y values above. It's not a BVM of course but it is a Sony CRT ...

yes, the two curves cross at 25%. I just used the functional fit from the Filmlight system documented earlier for the 1st column, and an average of that with a flat 2.4 curve for the second column. I tried the average function on my display and it was interesting, definitely a CRT look but too "metallic" - and definitely too dark for my display.

The BT.1886 rec. essentially answers the question of how to best emulate legacy CRT (which is highly variable) given your own display constraints.

Of the primary equation it says:
Quote:
 The EOTF specified in Annex 1 is considered to be a satisfactory, but not exact, match to the characteristic of an actual CRT
If for some reason you need even better emulation they recommend the alternate equation.
Quote:
 When it is necessary to more precisely match a flat panel display characteristic to a CRT, the alternative EOTF formulation specified below may provide a solution.
Attached is a calculator you can use to find target Y values for either function. btw, the primary function and secondary function overlap each other pretty well when primary black level=0.02 cd/m^2 and secondary black level @1.8%=0.1 cd/m^2, 100 cd/m^2 peak white.

Quote:
Originally Posted by zoyd

yes, the two curves cross at 25%. I just used the functional fit from the Filmlight system documented earlier for the 1st column, and an average of that with a flat 2.4 curve for the second column. I tried the average function on my display and it was interesting, definitely a CRT look but too "metallic" - and definitely too dark for my display.

The BT.1886 rec. essentially answers the question of how to best emulate legacy CRT (which is highly variable) given your own display constraints.

Of the primary equation it says:

If for some reason you need even better emulation they recommend the alternate equation.

Attached is a calculator you can use to find target Y values for either function. btw, the primary function and secondary function overlap each other pretty well when primary black level=0.02 cd/m^2 and secondary black level @1.8%=0.1 cd/m^2, 100 cm/m^2 peak white.

Thanks, Zoyd. Given that I have the same TV as you, you have tweaked yours for better blacks, and you didn't like the look if these settings, it appears you have saved me a lot if wasted time. So thanks TWICE.
Back to color science - I found this plot very interesting. What happens when you feed a linear BT.709 encoded signal into a legacy CRT with 0.1 cd/m^2 black lift?

The green line uses the alternate CRT matching function with 0.1 cd/m^2 black level. I don't think it's coincidence that the end-to-end gamma here is amazingly close to the dim surround recommendation from Hunt. I think those ITU guys are pretty smart.

Quote:
Originally Posted by zoyd

Why weren't encoding transfer functions designed to better match CRT's? That would have made the world a tidier place.

Maybe they were?
Quote:
Originally Posted by sotti

Yes BT.1886 is in CalMAN 5.

When CalMAN 5 comes out, will it make sense to calibrate to this new standard or is it really there for future use? Does all current program material like HDTV programming and BD movies use power law gamma?
New Posts  All Forums:Forum Nav:
Return Home
Back to Forum: Display Calibration
• How power law gamma calibration can lead to crushed blacks

AVS Top Picks

AVS › AVS Forum › Display Devices › Display Calibration › How power law gamma calibration can lead to crushed blacks