Grayscale variances - different patterns, plasma calibration - Page 6

It seems to me that we need to be more precise when we talk about gamma with respect to plasma and ABL. I think of it in terms of static gamma and dynamic gamma.

At any single moment in time, the image on the screen has more and less luminous areas. Static gamma is the input stimulus versus output luminance curve produced by the display at that moment in time. Because the effect of ABL at that instant is constant, it seems to me to be perfectly calibratable and the truly important definition of gamma from an image quality standpoint.

There is also dynamic gamma which is the curve produced (in the extreme case) by increasing from a fully black screen to a fully white screen. This is the gamma we can't truly calibrate because the amount of ABL in effect changes over time and is also affected by the size of the patterns used in performing the calibration. I just don't see this as being as important to image quality as static gamma. After all, when I'm outside and the sun goes behind a cloud for a moment and the world's ABL kicks in I don't start complaining about how bad the scenery looks.

My conclusion is therefore that I want my gamma to be flat when measured at a single level of ABL effect. If I can pick windows that are small enough that ABL has no (or almost no) effect, or if I use APL windows, and achieve a nice flat gamma I think that's the best quality picture. Of course, if ABL isn't implemented perfectly, which I would expect to be the case, things might drift for different levels of ABL activity. So I conjecture that it would be best to calibrate gamma using APL windows where the APL matches the average image luminance of normal content. Much like the idea of using 75% saturated colors when calibrating CMS.

The problem with the normal way we calibrate gamma, using ordinary windows, is that the amount of ABL effect is not constant and increases as we get to the higher stimulus levels. The gamma calibration is being performed on a moving target.
Roy

I completely agree. Here is a plot of the effect you describe, gamma vs. APL for a samsung plasma using a variety of pattern types. Do not be confused by the x-axis, it is NOT the pattern stimulus, it is the APL stimulus. In other words, each point on the graph is the average of a 10 pt. gamma measurement at a constant APL.

Notice that it is (relatively) constant from 20%-50% input stimulus. I've settled on using 22% APL patterns based on statistics of a cross-section of movies that Tom Huffman analyzed. Along with using the BT.1886 recommendation to tailor the shape of the gamma function, the results are very satisfying on this display.



Edited by zoyd - 2/9/13 at 8:05am

Could you please explain in more detail what your graph represents?

It seems like you're saying that the X-axis represents an average for all the video levels in multiple images used to measure a 10 point gamma run. If that is the case, in terms of how this relates to typical video content, the lower numbers to the left would appear to indicate darker scenes and the higher numbers to the right would seem to indicate brighter scenes. The graph would generally seem to indicate how on-screen gamma (rmongiovi's "static gamma") changes depending on the image displayed.

What I'm mainly confused about are the labels on the right. For example, what does "40% APL patterns" or "10pt off" mean? Another way to ask the question, how do the labels on the right relate to the actual patterns that were measured? Basically I don't understand how you can measure "40% APL patterns" and then also vary "APL (stimulus)". I'm guessing maybe you used a portion of the screen for the measurement area and then changed the background to vary "APL (stimulus)", like some of the measurements a long time ago. Currently I can't say exactly what the graph represents, because I don't understand the labels at the right, but it's zoomed in rather tight, so maybe the gamma variation less than 0.1 doesn't really matter.

One complication for me in these sorts of discussions is trying to determine exactly what people mean when they use the APL abbreviation. The AVS HD 709 grayscale measurement patterns marked APL hold a constant average input (video levels) for a measurement run, and they also hold a constant average brightness at the screen (gamma-adjusted image) during the measurement series to look at on-screen gamma. Patterns from other sources are likewise marked as APL and apparently intend to hold a constant average input (video levels) during the series of gamma measurements, but the average brightness at the screen (gamma-adjusted image) varies during their measurement runs and doesn't necessarily correlate with on-screen gamma. Depending on the individual display in question this distinction regarding APL patterns may or may not be relevant, but from my focus on trying to determine how well measurements from various displays actually correlate, the information from Chad B (http://www.avsforum.com/t/1446386/abl-effects-measured-for-comparison) seems to suggest it's probably relevant even when only focusing on plasma.
Edited by alluringreality - 2/9/13 at 8:04am

Sorry I didn't explain the legend correctly. Each curve was generated using the same set of APL patterns, one set for each x-axis point at the level indicated (1% windows on a constant background). This was done for 3 different settings of the 10pt control where gamma was set to 2.3 using the patterns indicated, and in the off position.

I'll just say wow. If commercial software did something like that it might be worthwhile, since it looks like Zoyd's graph came from at least 390 measurements. Personally I find the graph interesting, yet I still tend to think the measurements from Chad B likely suggest that pattern selection is probably the main factor in trying to have some correlation between gamma measurements on various displays, which can have their own idiosyncrasies in how exactly the display converts the input signal into light output that people measure at the screen. It's a bit puzzling how there's still some separation between Zoyd's settings that have been "calibrated" to a similar target, but it's only around a 0.05 gamma, which would appear to be smaller than the sorts of differences exhibited in gamma on similar measurements from Chad B's post if they too had been averaged to a single value. Of course the two comparisons use different displays, so it's difficult to say how similarly or differently the information might appear if it was all from the same display or if each post followed the same testing procedure.
Edited by alluringreality - 2/9/13 at 9:54am

I also did measurements similar to Chad B's and the shifts were larger (up to 0.1) when looking at smaller windows on a black background where APL is generally very low throughout the sequence, which is consistent with the graph above. The advantage to the plot above is you can immediately see over what range of input stimuli a gamma calibration will be stable for a given set of patterns and also which get you closer to your desired target all in one graph. Another thing you'll notice is that below 20% APL, gamma is changing just as fast as above 50% (where ABL is present). So the same rationale used to argue against using large patterns on plasmas applies to small patterns.

No, the above graph consists of measurements made only with 1% windows on a constant background. The background level was set to the values on the x-axis for each run. This was done for four settings of the 10pt control:

1. off
2. calibrated to a flat 2.3 gamma using 40% APL patterns
3. calibrated to a flat 2.3 gamma using 15% windows
4. calibrated to a flat 2.3 gamma using 27% APL patterns

The graph tells you that:

The display gamma varies with APL in a reproducible way, it increases from 5-20%, is relatively flat from 20-50%, and decreases from 50%-70%

So, if you want to avoid calibrating a "moving target" do as Roy suggested and pick APL patterns in the most stable range (20%-50%). I've chosen the low end of that range because the mean value of a number of movies analyzed by Tom ended up around 25%. Furthermore, the distribution function of the APL of those movies was asymmetric and skewed towards higher APL, so if you calibrate the display gamma set point to this APL the vast majority of images will have "static" gamma at the value obtained during calibration.

I agree in principal (everyone should try this measurement ) but the shape of that curve has been found to be consistent across different plasma brands and a wide variety of patterns. The details of where the upper ABL induced cut-on occurs will vary based primarily on the contrast setting but there is consistent behavior that a stable region exists and fortunately matches up with typical real APL averages (maybe by design?) so I see no reason not to use patterns the measurements are telling me are most appropriate.

By the way, if you have a plasma display and want to test whether or not the gamma is varying at low APL as shown above, take a look at the dynamic contrast test on the AVSHD disk. If you see the lower end of the grayscale steps get brighter as the overall picture level gets darker, then your plasma would measure a similar functional shape as mine.

You still don't have it. Here is the procedure I followed:

1. no adjustment of the 10pt control (off)

a. Measure the display gamma using 1% windows on a 5% background to derive the 5% x-axis point
b. Measure the display gamma using 1% windows on a 10% background to derive the 10% x-axis point
c. Measure the display gamma using 1% windows on a 10% background to derive the 15% x-axis point
d. Measure the display gamma using 1% windows on a 20% background to derive the 20% x-axis point
e. Measure the display gamma using 1% windows on a 30% background to derive the 30% x-axis point
f. Measure the display gamma using 1% windows on a 40% background to derive the 40% x-axis point
g. Measure the display gamma using 1% windows on a 50% background to derive the 50% x-axis point
h. Measure the display gamma using 1% windows on a 60% background to derive the 60% x-axis point
i. Measure the display gamma using 1% windows on a 70% background to derive the 70% x-axis point

2. Adjust the 10pt control to calibrate the display to a flat 2.3 gamma using 1% windows on a 40% background.

a. Measure the display gamma using 1% windows on a 5% background to derive the 5% x-axis point
b. Measure the display gamma using 1% windows on a 10% background to derive the 10% x-axis point
c. Measure the display gamma using 1% windows on a 10% background to derive the 10% x-axis point
d. Measure the display gamma using 1% windows on a 20% background to derive the 20% x-axis point
e. Measure the display gamma using 1% windows on a 30% background to derive the 30% x-axis point
f. Measure the display gamma using 1% windows on a 40% background to derive the 40% x-axis point
g. Measure the display gamma using 1% windows on a 50% background to derive the 50% x-axis point
h. Measure the display gamma using 1% windows on a 60% background to derive the 60% x-axis point
i. Measure the display gamma using 1% windows on a 70% background to derive the 70% x-axis point


...rinse/repeat starting with 15% windows and 27% APL

9x11x4=396 measurements

no, the 27% and 40% APL were 1% windows on fixed 27% and 40% backgrounds and again these were just used to set the 10pt controls for flat 2.3 gamma not to generate the plot. Each point on the plot is the average gamma for 1 grayscale run at the specified APL using fixed background patterns.

Basically Zoyd is using a 1% window for all the "APL" measurements, but instead of using a black background he's using the background to set average video level. There is a bit of change in his "APL" measurement series, but I would consider it rather minimal, especially since he's averaging the entire gamma run to one value. Between a pattern to measure black and a pattern to measure white, the average of the video levels on his patterns change by the window size of 1%, and a gamma-adjusted average level expected at the screen also only changes by 1%. Even on displays that are the most sensitive to changes in screen brightness, since Zoyd is averaging gamma to a single value, I would generally expect to see minimal difference in gamma measurements against a pattern that truly holds a constant screen brightness during the gamma measurement run.

To test that theory, I compared the "2% window" gamma graph against the "DB Ramp Dark" gamma graph in Chad B's PDF (http://www.avsforum.com/t/1446386/abl-effects-measured-for-comparison). They are generally both very dark measurement series, as shown in their similarly high "Y Max fL" numbers. The "2% window" measurements have twice the variation in average video level and gamma-adjusted average level when compared to Zoyd's patterns, because the windows have twice the area. On the other hand the "DB Ramp Dark" has zero change in average video level or gamma-adjusted average level for the gamma measurement series. The "2% window" and "DB Ramp Dark" ramp graphs share a similar shape, and if the gamma measurements were each averaged to a single value there would be little difference between the "2% window" and "DB Ramp Dark" ramp gamma measurements. If Chad B had done a 1% window measurement, which is the measurement area size Zoyd is using, I would suggest there probably would have been even less variation against the "DB Ramp Dark" gamma, especially when each gamma graph was averaged to a single value. My main point here is to say that Zoyd's patterns likely measure gamma very close to using any type of measurement that holds a constant on-screen image - grayscale bars, AVS HD 709 APL patterns, Chad B's APL (http://www.avsforum.com/t/1445226/chad-bs-custom-test-patterns), etc.

I cannot say how exactly some publicly-available patterns like Chad B's APL (http://www.avsforum.com/t/1445226/chad-bs-custom-test-patterns), "AVS S APL", or "AVS L APL" would relate to Zoyd's measurements, but I would suggest that their gamma measurements likely have some correlation to the measurements that Zoyd is taking. I get the impression that Zoyd thinks average video level is most relevant when it comes to plasma. My opinion is highly based on other display types, which can clearly vary light output while average video level remains constant, so I like to also look at ideal gamma-adjusted levels at the screen. From the data I've seen on plasma I get the impression that the information doesn't strictly tie average video levels to how the display operates, so I tend to think of patterns that only try to hold average video level constant as likely no better for gamma measurements than typical windows, even on plasma. Like I was basically saying in post #153, all "APL" patterns are not created equal, but patterns like the following probably tend to share some similarities for gamma measurements.

Publicly-available gray patterns for constant on-screen measurements:
AVS L APL - 36% average video level or 25% gamma-adjusted average level
Chad B's APL - 19% average video level or 13% gamma-adjusted average level
AVS S APL - 7% average video level or 5% gamma-adjusted average level

Zoyd's patterns that probably exhibit similarities to constant on-screen measurements:
40% APL - 40% average video level or 13% gamma-adjusted average level
27% APL - 27% average video level or 6% gamma-adjusted average level
22% APL - 22% average video level or 4% gamma-adjusted average level

Note: The levels listed here are all approximations, and 2.2 gamma was used for the gamma-adjusted approximation


My guess is that I suspect Zoyd's "22% APL" is probably darker than a lot of typical video content, but it would be interesting to at least see how the individual gamma measurements for his "22% APL" and "40% APL" patterns compare against Chad B's APL. The reason I would like to see a comparison of these three patterns is because the Chad B's APL has an average video level that isn't too different from Zoyd's "22% APL", while the gamma-adjusted average level for Chad B's APL is similar to Zoyd's "40% APL". I'm not sure if either average video level or gamma-adjusted average level can necessarily predict display performance, so I would guess gamma on Chad B's APL might fall somewhere between Zoyd's "22% APL" and "40% APL" patterns. My guesses have turned up to be completely wrong at various times in the past, but I would consider it interesting to see how gamma on Chad B's APL compares against Zoyd's "22% APL" and "40% APL" patterns on one of these two displays with the most posted information.
Edited by alluringreality - 2/10/13 at 1:09pm

That's what I thought too until I saw these measurements . The ensemble average of that dataset is 22% +/- 4% 1-sigma. In reality the gamma shifts I measure when using 22% vs. 40% APL for example are very small and even 15% windows shift < 0.05 relative to these. But once you start using smaller non-APL patterns, say in the 1%-10% by area range, which is typically recommended for plasma, the shifts can approach 0.1 gamma. It's a noticeable effect at that point, leading to crushed blacks in typical APL scenes.

I'll assume APL on the graph refers to average video level. If the same information was represented in another way, such as an expected gamma-adjusted average level at the screen, there would be a different "ensemble average of that dataset". I'm just not sure what measurement data strictly relates average video level with plasma display operation, so the suggested measurement comparison was sort of an extension to your power draw test from 2011 (http://www.avsforum.com/t/948496/avs-hd-709-blu-ray-mp4-calibration/3000#post_21039129).
Edited by alluringreality - 2/10/13 at 2:25pm

Looking around for some more information I ran across what seems to be some definitions from Charles Poynton, which also appear to support the APL definition from the Video Essentials website. What I've been calling "video level" seems to typically be termed "picture level", and what I've previously listed as "gamma-adjusted average" appears to be correctly termed "average relative luminance". So average picture level (APL) simply refers to an average of the video information, while I personally tend to think it might make sense to also at least look at what is going on with average relative luminance. Anyway as it applies to a gamma discussion, I believe Charles Poynton would term the measurements you take at the screen luminance, or linear light, or Y. He would refer to the non-linear video information as luma, or picture level, or Y'. Generally luminance and luma at the display is expected to be related by a power function, or display gamma, but this isn't exactly what you measure when using typical window patterns.
Edited by alluringreality - 2/10/13 at 3:15pm

What should be set the brightness and contrast on your TV ?
brightness - 50% ?
contrast - 100% ?

Or another?

Hello,

Here's my gamma curve I've got after calibrating my RGB's.
It hasn't changed much from what I've got first (before calibration).


P65ST50 - Small APL patterns from AVSHD - i1D3 PRO - Chromapure

I noticed that my gamma curve was flat when I took my measurements with the "PLAY" logo (from my BD player) showing on the screen.
So be aware that this small element can and will influence your gamma curve.

Light Illusion
Why can not measure 50% IRE full field contact method and get the result desired patch size for the profiling ?
Or average sum 0-17-34-51-68-85-102-119-136-153-170-187-204-221-238-255 IRE full fields ?

Whether the difference with non-contact method ?

Sorry, I don't understand the question?

What are you trying to suggest???

Steve

Sorry, just re-red my post and it sounds too blunt - I didn't means that!
I am just not sure what you are trying to suggest.
If you can explain a bit more I'll see if I can answer.
Edited by Light Illusion - 2/21/13 at 1:08pm

Light Illusion
Thank you and excuse me for my questions.

target - use the contact method of sensor.

I suggest:
measure
1) 50 IRE full field and then make a calibration patch window with 50% white and sized it to get a similar Luma reading with the probe in the same place
2) 0,17,34,51,68,85,102,119,136,153,170,187,204,221,238,255 RGB patterns - average sum - and then make a calibration patch window with 50% white and sized it to get a similar Luma reading with the probe in the same place

Sorry, I am still confused.

I can't see how what you are suggesting would work?
The issue we are trying to overcome is plasma's ABL.
I can't see how your suggestion would work?

Sorry - I think I'm still missing something.

I don't think so...

@anta1974 - your suggestions indicate that you do not have a proper understanding of the ABL issue.

Light Illusion proposes to measure Grey-scale Ramp image for overall brightness of the screen non-contact probe.
I propose to measure parts of Grey-scale Ramp image contact probe.
I believe that by measuring the full field of 0/0/0 to 255/255/255 and calculating the average of sum I get the desired brightness only measured contact method.
ABL will work on the complete fields individually and as on-ramp!

And I see you have a proper understanding of the ABL issue, then perhaps be able to offer a way of finding the right size pattern contact method? Or is it impossible?

What anta1974 is asking is actually quite simple...

Measuring a grayscale ramp pattern, full-screen, to get a luminance to determine how to setup your measurement patterns is REALLY REALLY REALLY difficult (and frankly, I'm surprised a calibration-related professional would even suggest trying to measure a full-screen ramp pattern at all because of the extreme level of difficulty in doing that with the tools we use for calibration). Many meters have no viewfinder or other indicator of exactly what the meter is measuring (other than results of measurements themselves). And most meters measure a circle, not a 16:9 rectangle. The meter I use DOES have a view finder and it only measures a 1-degree field of view when set to the largest viewing angle it supports. I would have to move the meter 250 feet (~80m) from a 60" diagonal flat panel TV for my meter's 1-degree viewing angle to encompass the entire screen of the flat-panel TV. But even if I could do that, it would still be measuring a circle larger than the TV screen. Other types of meters have very large angle-of-view and might not need to be very far from the TV at all -- but it would be difficult to know if you were measuring 100% of the TV screen or 90% or 80% or even if you were so far away that you were measuring 200% of the screen area (the screen plus a lot of black area). Most meters' viewing area is circular while the plasma screen is rectangular so you can't really get JUST the ramp pattern without excess black area surrounding the screen -- making your luminance measurement lower than what the TV is actually displaying.

So... the entire point for anta1974 asking his questions is because of the difficulty we have trying to measure the luminance of the ramp pattern properly with the tools we are all using for calibration.

His 2 proposals are for ways to measure the light output of the panel in some static state that won't cause the ABL from influencing the luminance measurement results.

His first suggestion is to use a full-screen pattern that produces 50% of the luminance of 100% white... in our world where we are (or should be) using gammas in the range of 2.2-2.3, a 75% white pattern would produce close to 50% of the luminance of 100% white, but it wouldn't be exactly 50%... you'd need a range patterns, like 74.0%, 74.1%, 74.2% etc. in order to find the one that's exactly 50% of the 100% white luminance. That would tell you when the ABL is in the middle of it's limiting range (presumably -- but who knows for sure, could even vary by brand/model/year). You could then use that luminance level as your target level for full-screen patterns that maintain the same luminance level (the central measurement area and the background area would vary as needed to maintain your target luminance level. The assumption here is that the grayscale ramp pattern luminance measurement would produce the same luminance measurement as a full-screen, single-level pattern that measured 50% of the luminance of 100% white.

Seems to me, something like that would be far more practical than trying to measure a full-screen grayscale ramp pattern. The question is whether it would produce the same result as measuring a grayscale ramp pattern.

For my meter to measure a full-screen grayscale ramp pattern using a meter that has a 1-degree viewing angle, if my plasma panel is setup so I have a 35-degree viewing angle and I can't move my meter any farther back than where I sit, I would have to take around 70 measurements across the screen (overlapping because of the circular viewing area of the meter) and average those measurements to get the average luminance of a grayscale ramp pattern. So the poster is looking for some alternate, more practical way to get the same result you get with a measurement of a full-screen ramp pattern... something that could be done with just 1 measurment or perhaps a small series of measurements that are do-able with the meters most of us are using.

Doug,

Great explanation. Here's what I have been playing with. It's not scientific scientific really, and maybe a bit situational, but seem to have a thin thread of logic Though I am likely as not missing a whole series of factors, which I hope you/somone can help with.

-put up a window (20% or less), measure light output
-put up a full field, measure light output

*If the greatest output is a window, then one could further go down in window size further (as they have patterns available) and see if there is a drop off or gain in light output again.

The reason I was thinking the above, is that I was doing a Sony the other day and the pre-cal gamma log/log didn't match the post-cal (pre was smooth and the post had a drop/dE spike at 90%). Maybe this is a completely different issue or concept, but please hang with me.

For some reason, the window size had been changed from full screen to a x% window patterns. So, once I noticed, I changed the patterns back to full screen and voila, fairly linear gamma log/log at 2.2, no more spike.

So, I was thinking, if window size impacted the gamma log/log linearity (and it wasn't a Sony and/or fluke of sorts), then maybe that might be something to work with to determine a good size pattern for minimizing ABL.