2.5 is display gamma, NOT 2.2 - Page 3

Reading Poynton a little closer reveals the answer. Rec 709 inverse transform gives scene tri-stimulus values, useful for intermediate processing. You need a straight 2.5 power law curve to decode to display tristimulus values and preserve rendering intent.

If it's good enough for Poynton, it's good enough for me. Anything gamma corrected with Rec 709 should use a 2.5 straight power law gamma curve on the display side. For older content encoded with an different transform, the correct display gamma may be different.

When you have the time please explain that to Sony, Ikegami and Dr. Raymond Soneira for that matter. Dr. Soneira actually measured the Sony PVM-20L5 to be ruler flat 2.20. Here:

http://www.extremetech.com/article2/...1736943,00.asp

http://www.extremetech.com/article2/...1736944,00.asp

Sony and Ikegami each designed their Pro reference series direct view CRT mointors (~$20K) with custom processing to achive the "Ideal" 2.2 display response. If 2.5 is the industry benchmark, then the additional processing (circuitry) engineered into these devices are broken by design according to Poynton?

That depends on how the data was gamma corrected in the first place.

From Poynton: For NTSC, the 1953 FCC statement says "... R'G'B' encoding for a display with a transfer gradient (gamma exponent) of 2.2".

This is commonly interpreted as a 2.2 display gamma to be used for NTSC content. The standard is poorly documented, and Poynton recommends using Rec 709 encoding and gamma 2.5 decoding for all types of video. Pre Rec 709, gamma 2.2 was considered the standard and I'm sure that many manufacturers still make this assumption.

There is really no good answer, and I don't want to get into a pissing match about what is the "correct" display gamma. If you watch a lot of HD content, encoded with Rec 709, go with gamma 2.5. If you watch a lot of older NTSC content, feel free to go with 2.2. Call up the content creators and ask them what they use for their reference monitors. I'm sure you'll get all kinds of answers (including "I don't know"). Some content looks so poor that I doubt grayscale is even calibrated correctly in the studio.

We are in a transition period, where Rec 709 transforms (and gamut) are reaching the production and broadcast houses. Hopefully the new standard will be in place across the entire chain in the years to come.

Awfully definitive for someone who knows there's no good answer. The principle is to match your display's gamma to your viewing environment. Is there a hard translation for "X lux = Y gamma"? Not that I've ever found, but I'm sure there is a lot of the SMPTE back-library which I haven't had a chance to research.

Bill

I don't think there is either. Didn't mean to be hard-line here.

I understand the benefit of using inverse 709 (2.22 approx) during intermediate processing, and perhaps this also includes studio work. Professional cameras tend to be more conservative with levels, generating a slightly washed out look. This is to avoid clipping and leave some headroom down the line for rounding errors etc. Maybe a studio monitor gamma of 2.2 is useful for the operator to see deeper into the blacks and make the workflow less of an eye-strain? It would reveal more of the picture than a 2.5 gamma monitor. Studio processing could be considers part of "intermediate processing" and should then follow inverse 709. I haven't found the specification for a studio monitor in any standard.

If this makes sense, the "final push" to 2.5 should only happen in the home, where a higher gamma (=more contrasty look and higher saturation) is used to deliver the "rendering intent".

This is of course pure speculation.

Bravo... and I think the quoted statement applies to all of the comments made about display gamma... in the face of lacking a definitive standard, comments about "implicite" assumptions... well, you know what they say about ASSumptions...

Imagine yourself in the shoes of the standard writers and how they reasoned.

Assumption 1: The standards were designed with CRT displays in mind
Assumption 2: Most consumer CRTs at the time had a native gamma response of ~2.5
Assumption 3: The writers of the NTSC and Rec 709 standards were aware of (1) and (2)
Assumption 4: They choose a gamma correction transform close to 2.2 despite (3)
Assumption 5: They did (4) because they wanted a system gamma > 1.0

I think Poynton is trying to say:

"If you design displays with a gamma 2.2 response, are not taking into account all the intentions of the standard writers, even if these are not specifically stated. By simply reversing the transform, you are missing an important, unstated, assumption."

I think it's a good argument.

PS I have left voicemail for a few broadcast engineers from my local stations. I'll report what they come back with. DS

O.k. then.

In HCFR, I matched my gamma to 2.22 with the display gamma reference. These measurements also give me a roughly 2.5 gamma using the "camera gamma" reference. So which should I be using? I have a Epson 1080UB with excellent black levels, so black crush isn't as much of an issue.

I'm going to try Calman, which is an excellent product, to compare against HCFR and see what differences there are for gamma.

Dan

What about the sRGB color space? Does it escape all this controversy with its established 2.2 display gamma?

Forgot to report back. The broadcast engineers I talked to all agreed that they had no idea what the gamma was on the monitors, and they never measured it. The only cared about black level, contrast, color, tint and white point.

Try this:
http://www.w3.org/Graphics/Color/sRGB.html

The entertaining thing is the goal of maintaining the total system gamma, when most office/computing environments are much better lit than home video/HT viewing environments. That being said, sRGB has created a LOT of controversy, this only being one of the issues (reduced gamut being an even bigger issue).

I think it says they went for a 2.2 display gamma that goes well with dim lighted rooms instead of sticking to 2.5 for completely dark rooms.

Makes sense as it's kinda impractical to assume that general computing will always be performed in a dark room. I've personally decided to adjust my computer screen for 2.1 as I pretty much always use it with the room lights on, so far so good. My office laptop I think I will set at 2.0, it's way too bright in there.


Thanks a lot for the link Bear

I think you are misinterpreting what is in the link. Here's my take:


Interpretation: "OSHA be damned. The work environment that most of this content is going to be used in sucks, so we hope they figure out that they need to turn off some lights, and we're going to write the standard in a way that's easy. Oh, yeah, here's a study that makes us feel good, too."


Interpretation: "We had an intern who was playing with OpenOffice one day, and he noticed that if you use a completely different formula for gamma than what is specified in Rec. 709, then you get a different coefficient than in Rec. 709. We didn't want him to feel bad about not getting paid, so we went with it."

I'm not sure it went down exactly like this, but you get the idea. Bruce Lindbloom has a better take on this (follow the links at the bottom of this page):
http://www.brucelindbloom.com/index....SpaceInfo.html

Hmmm, I think you are saying that they should have gonne with a gamma standard aimed at 200 lux rooms. Maybe they didn't want to deviate that much from HDTV standards especially when TV and PC screens are pretty much the same these days. I wonder how many lux my room has with the shades closed and the lights on.


Thanks for the link, I'll read it when I'm at work (mehehe).

The actual issue of course is that when speaking of multiple gammas you have to use a power function if you want to use the simple math, for example in calculating viewing gamma (end-to-end gamma). Poynton chooses to completely ignore the offset defined in the Rec 709 gamma correction (Section 9 from http://www.poynton.com/notes/colour_.../GammaFAQ.html shows the function) and just use the .45 exponent in simplification, although in the online documents he really gives no reasoning why he ignores the linear portion for simplification. The sRGB document uses an exponent of 1/1.956 for 709 in the same way as the section "A note about video gamma" from http://www.w3.org/TR/PNG-GammaAppendix.html to better account for the linear portion. What happens is that there is a larger offset near where the linear portion exists if you use .45 instead of 1/1.956 for the simplification.

This same sort of item also appears with sRGB gamma because, like with the 709 gamma correction, it isn't a simple power function. The brucelindbloom link says of sRGB "Note: The gamma of sRGB is not exactly 2.2, but rather, is a grafting together of two different functions, that when viewed together, may be approximated by a simple 2.2 gamma curve." Rec 709 also uses two functions in defining gamma correction, and the simplification of 1/1.956 is more in line with the use of the 2.2 gamma for sRGB. In making this comparison it should be noted that the function for E' from section 9 in http://www.poynton.com/notes/colour_...mma_correction is very similar to the function for R in http://www.brucelindbloom.com/index....YZ_to_RGB.html Someone would have to explain to me why it makes sense to simplify one using the exponent for the major portion of the function and then not to do the same with the other.

At the display end where we all work, the linear portion of the response curve is of particular concern. Poynton seems more concerned with the overall concept and brevity. This, and every other discussion of gamma that is of any value, seems to be moving toward a recognition that the issue is far more comples than whether 2.2 or 2.5 is the right number. The fact is, a single number does not provide adequate information about the response of a display. You have to look at the curve. A single value can be very misleading. It is like trying to talk about the color of white using color temperature, only worse.

Shouldn't gamma be set somewhere between 2.2 and 2.5, but ultimately be set based upon the good old eyeball.....so that you have the most depth and pop without sacrificing shadow details?

Gamma should be adjusted to track as closely to the response curve appropraite for the application, in most cases here that will be rec709 response. The single value that you get will depend on the display, the calculation, and the variance with respect to the reference chosen. Many displays only allow very crude adjustment. The relationship between proper tracking and the value calculated can vary greatly. The bottom line is, approximate the curve, then tweak for the needs of the system. Yes, the final adjustment is done visually.

The average slope for the standard Rec. 709 power function (aka gamma) has a value of 1/0.45 (2.22). Actually this curve has an approximate value (or average slope in the linear midtone region) closer to 1/0.52 (1.92) after the grafted on linear segment near black is considered.

Most professionals capturing content do NOT simply use this default Rec. 709 curve or tonal pre-emphasis, they typically modify the cameras midtone or gamma response as well as adjusting reference (balance) white, flare compensation, master pedestal.. along with compression thresholds such as black stretch and knee contours.

Amen.

I take a little pride in our tools for this purpose, even though in the not-distant-enough past I used to be a lot more dogmatic on the subject of the Rec. 709 transfer function. The area where I am currently dogmatic about is stamping out the "one number" fascination with gamma. I have seen systems that have a "perfect" 2.2 - 2.5 gamma (pick a number), which have HUGE issues with grayscale (crushed blacks, gray blacks, crushed whites, etc.). No matter which calibration tool you use, you need to think of gamma as a target that you have to hit multiple times, not just once and on average, to get a pleasing image in your environment.


I am convinced that they did not want to deviate from 709, since part of the point of sRGB was to bring 709 to the PC universe. The issue is more of one of how one sets a "standard". For a standard to be credible, it must be a) realistic, and b) provide a collective experience that everyone can agree gets most of the major things right (i.e., people can nitpick at the margins, but not have widescale revolt).

sRGB, from what I can tell, fails at both of these. It is not realistic in that one of the major design goals -- colorspace for the web -- does not seem to match peoples viewing environments or content development environments (often, brightly lit office spaces). It also has significant comprises versus existing, de facto standards (e.g., Adobe RGB) that make it too big of a compromise for the scale of its total ambition (e.g., end-to-end unification of web publishing and print publishing). Given this, and the fragmented market it seeks to serve, it's no wonder that it is having trouble gaining traction compared to, say, the Broadcast and Film industries with 709.

Agree.

Have you ever calibrated a display to perfectly or even nearly track Rec 709 response?

I have found that a simple power function is the only practical response for display gamma. Attached is how the same measurement data looks like when using the two different formulas (yellow line is the average, red line is a least squares fit) for a rough calibration of Pioneer 428XD plasma TV, A/V Mode ISF Day, using CalMAN 3.0.1, ControlCAL beta 2 and a Display LT colorimeter.

Here are the measurement data:

Looks awfully difficult to get this display to track Rec 709 response. It would be very interesting to see measurement data from a practical example with a display tracking the Rec 709 response curve.

Or have I misunderstood something here?

PS This is of course ignoring the linear part near black. Is this the problem with the approximation?

Real displays should not include or reciprocate the linear tail response.

The linear region helps reduce noise by reducing gain (production end processing) near black.

CRT approximates a power function response, so it's a reasonable assumption that a power function is the intent for display gamma with Rec 709. It seems to me that what is really being accounted for with display gamma is the perceptual issue commented on in "Surround effect" from http://www.poynton.com/PDFs/TIDV/Gamma.pdf or "What should overall gamma be?" from http://www.w3.org/TR/PNG-GammaAppendix.html In either explanation the point is that to account for dim viewing conditions display gamma should result in a more contrasty image than the original. This more contrasty image comes from an end-to-end (viewing) gamma greater than 1 as commented in either document.

What is actually defined in Rec 709 is gamma correction (Section 9 from http://www.poynton.com/notes/colour_...mma_correction). Typically in a discussion of this sort the gamma correction function is simplified to a power function so that end-to-end gamma can be discussed. Because the original function includes the linear tail, it makes most sense to me to include that portion in the simplification. I would basically consider an approximate power function nearer 0.51 more reasonable than 0.45 for gamma correction, but this is generally discussed in my prior reply.

The inverse of the simplified gamma correction would be an end-to-end power function (gamma) of 1, but for a dim environment the target is greater than 1 as previously commented. The Poynton document references a range of 1.1 to 1.2 for end-to-end (viewing) gamma, but I've also seen 1.25 referenced more than once. Generally the trend seems to be that higher end-to-end gammas are recommended for darker environments, but another thread on this subject generally didn't put much emphasis on the effect of ambient light between a "dim" and "dark" room. If you have a simplified gamma correction and a target end-to-end gamma then the following applies:

end-to-end gamma * 1/(gamma correction) = target display gamma
For example: 1.25 * 1/0.51 = 2.45 display gamma
Another example: 1.10 * 1/0.51 = 2.16 display gamma

Of course in the reduction of display gamma to a single target, it would be best to look at the display gamma (power function) across the grayscale, like in your image, rather than looking at an averaged display gamma.

wow, I've been reading this thread for the past couple of days and consider myself pretty techie (I've been in IT for 11+yrs) and this whole gamma thing is going right over my head...LOL

good thread though, I think I am learning from it as well..HAHA

Don't feel bad man, I think we IT guys are just hard wired against analog stuff and solving problems with other problems. XD

Gamma has been adapted to serve many roles simultaneously, obscuring its original purpose. The primary considerations in the creation and calibration of displays are to minimize quantization on a lossy (8 bit) channel, to linearize the display response, and to account for the change in perception of the observer to maintain rendering intent.

???