Does a gamma measurement provide comparable information?

Calibration theory suggests grayscale gamma is a primary consideration in trying to make various displays look similar using measurements. One complication when relating theory to actual practice are the numerous types of displays, which each have a number of display settings. The various available displays and settings often tend to provide different gamma measurements, depending on the selected measurement pattern. The main question I’ve had for years - does a gamma measurement provide information that can be compared with other displays or settings? Anyone is welcome to follow along and correct where my impressions may fail to fit with available data.

The general forum consensus seems to suggest consistent gamma measurements require different measurement patterns, depending on the display or settings. The main exception are the displays that can return similar measurements regardless of the pattern, like fixed-backlight or fixed-iris models. If you happen to have one of the displays that provide consistent gamma measurements then apparently your information is relevant, but rarely do I read about people making side-by-side visual checks to see if similar measurements on other displays actually compare well against displays with predictable gamma measurements. Essentially the general forum advice seems to be that calibrators need to assemble a compendium of the differing opinion regarding which measurement pattern should be used with each type of display and learn how to apply the information. Personally I cannot remember seeing measurement data to support the suggestion that the easiest way to arrive at similar gamma measurement information is by using dissimilar measurement patterns, so at best I consider the general opinion regarding gamma measurements as difficult to implement for a typical person using an unfamiliar display. Generally I expect that gamma measurements probably often result in limited correlation among various displays, which just means similar gamma measurements may actually result in dissimilar display performance.

Based on the measurements I have seen, it appears that various displays and settings primarily operate differently depending on how information is presented to the display, and I do not see a reason that unique measurement patterns are necessarily required for generally comparable gamma measurements on a majority of displays. Certainly if you only want to take a single gamma measurement run it makes sense to carefully consider how to assemble a gamma measurement pattern to meet your intentions, but I see no clear reason to doubt if holding a constant average picture level (APL) and a constant average luminance (AL) during a measurement run provides a single point of comparable gamma information across various displays and settings. Personally I have not measured patterns that only intend to hold a constant average luminance (AL), yet based on the preliminary notes for the upcoming Spears and Munsil disc, it would appear that everyone is not tied to the idea that various displays require dissimilar measurement patterns for gamma.

By this point I don’t know if anyone is still following along, but I’ll explain how I view some of the plasma gamma measurement data posted in the last few months. Since I don’t have access to the plasma displays with an opportunity to make subjective comparisons, like viewing the plasma display side by side with an LCD that has a fixed-backlight, I choose to limit which data I consider most relevant. In my own measurements on non-plasma displays, gamma measurements that hold APL & AL constant seem to have the best correlation between gamma measurement data and my subjective impression when compared to how an LCD-based display with a fixed-backlight would measure. Some displays can clearly vary relative gray measurements as APL or AL changes, so if APL & AL are constant during the measurement run there are simply fewer variables to possibly influence the reported gamma information on displays that do not provide consistent measurements regardless of pattern.

Primarily I’m just going to discuss the plasma measurements from Chad B (See linked thread). The items he measured that hold a constant APL & AL during the measurement run are listed below from darkest to brightest (See linked PDF for graphs):

- DB Ramp Dark (Darkest image)(4% APL & 3% AL)
- AVS S APL (Dark image)(7% APL & 5% AL)
- Chad B’s APL (Typical image)(19% APL & 13% AL)
- AVS L APL (Bright image)(36% APL & 25% AL)
- DB Ramp Bright (Extremely bright image)(85% APL & 84% AL)

When I look at the listed gamma graphs I compare them to each other for general shape and how they are located vertically in relation to the horizontal reference line. Generally most of the listed gamma graphs approximate to a rising line between 20% and 60% gray, and the graphs are generally above the colored horizontal reference line. There are a couple anomalies on the brighter graphs, where the AVS L APL has a dip at 80% gray and the DB Ramp Bright has a greater dip at 90% gray. I don’t know if I would be able to easily notice possible changes on brighter scenes with typical video content, but I suspect that I would be able to notice how the display changes gamma between dark and bright scenes in a dim room, which is the increase of approximately 0.2 gamma for 20%-60% gray when comparing the DB Ramp Dark graph to the AVS L APL graph.

Based on the measurements from Chad B, I would assume that when his plasma is calibrated using the 21% APL patterns (closest to reference) it causes the on-screen image (5 patterns listed above) to generally have a higher gamma than the 21% APL measurement indicates. Chad B’s APL and AVS L APL measurements are the most above reference, so I would expect typical scenes and bright scenes to have the most pronounced difference compared to an LCD calibrated with the same 21% APL pattern. Personally I doubt if using the 21% APL pattern to calibrate both the given plasma and an LCD with a fixed-backlight would tend to make identical gamma measurements result in similarly-looking images if the two displays were viewed side-by-side, since the on-screen measurements (5 patterns listed above) suggest that the plasma might tend to actually have a higher gamma.

If the five listed graphs from Chad B were each averaged to a single value, they would appear to fit with how Zoyd has documented the way his plasma changes average gamma depending on the brightness of the image displayed (See link). The measurements from Chad B and the information from Zoyd seem to suggest that on-screen gamma changes on their plasmas to some extent depending on the image displayed. The information from Zoyd also suggests that the variation on his plasma is somewhat independent from how the display was calibrated, since the graphs for both the 15% windows and his 1% area patterns generally parallel each other. I have read suggestions on this forum that historically windows have been used for gamma measurements because they allow a user to calibrate for a wide range of image brightness, but the information from Zoyd or Chad B doesn’t appear to support such suggestions, since for example the on-screen measurements from Chad B do not appear to generally relate with the window measurements. Unfortunately I don’t believe current consumer-level software can easily provide the amount of display-specific data provided in Zoyd’s graph, and in order to similarly document display performance would likely require a lot of time or automation of some sort.

I cannot come to the sort of one or two sentence conclusion that the average forum reader will likely want, regarding how they should attempt to measure and calibrate gamma. My own interest was mainly to see if anyone has additional factual data that might suggest how well or poorly gamma measurements may coincide. I am aware of deficiencies unrelated to gamma for displays that I have tried measuring closely, so maybe it should not be shocking to have measurements suggesting that modern plasma might vary on-screen gamma by a 0.1-0.2 range somewhat regardless of how the display is measured and calibration attempted. For the people that invariably want to know what single measurement series they should use to calibrate their display, I still get the impression that multiple measurement runs can give more compete information about display performance. I consider it unfortunate that measurement software does not appear headed in that direction, so I'll hope that the idea behind the measurement patterns on the future Spears and Munsil release provides better correlation for gamma measurements across various displays and settings than my impression of the current situation.
Edited by alluringreality - 3/14/13 at 11:55pm

To make what I wrote more accessible, I probably should have addressed the question - what is display gamma? I'll give a summary of grayscale gamma from my perspective for someone that has a video source (ex. Blu-ray player) and intends to calibrate a display, and for more information maybe take a look at the FAQ from Charles Poynton or the gamma correction wiki.

Essentially gamma simply defines how bright one displayed video level should be in relation to other video levels. Lets say that you only display two video levels at the same time, a 50% gray and 100% white. Gamma simply defines how bright 50% gray should be in relation to the 100% white when they are both on screen. Many times people talk about things like a 2.2 or 2.4 display gamma. In this situation, where we simply have two video levels on screen, the gamma number merely defines how bright 50% gray should be compared to white. If you were targeting a 2.2 gamma then you would want 50% gray to be approximately 21.8% of the brightness of white, and if you were targeting a 2.4 gamma then 50% gray should be approximately 18.9% of the brightness of white. Primarily gamma intends to merely state how the brightness of displayed video levels relate to each other.

To display a more complex image there would be additional video levels on the screen at the same time. Blu-ray gives only 220 gray video levels from black (16) through white (235). If we had a perfect display then gamma would define the intended brightness for each of the 218 video levels between black and white. Typically gamma is calculated relative to white, so if white changes then attempting to maintain a constant gamma value requires all the 218 levels between black and white to change in concert with white. Regardless if we are looking at 2 or 220 gray video levels, the basic intent behind gamma is to define relative brightness for displayed video levels.

Some LCDs and projectors provide easy to understand gamma measurement information in line with the ideal manner previously described, since each video level including white may have a somewhat constant measurable brightness. On these sorts of fixed-backlight or fixed-iris displays, where white remains constant, you can generally expect the measured gamma information to relate with the image you see displayed on your TV. The projectors and LCD-based displays I'm discussing here often have actual uniformity issues, where different areas of the screen may vary brightness in relation to other areas of the screen. For example, projectors generally tend to be brighter in the center than at the edges, and LCD-based displays may exhibit "clouding". Generally uniformity issues do not affect gamma measurements, because calibrators commonly take measurements from a single location. but ultimately uniformity issues mean that a single measurement series does not fully describe how the display actually operates and exactly what you see on screen. Since calibrators commonly measure gamma from only one area of the screen, some LCDs and projectors measure closely with the theory behind gamma, regardless of the measurement pattern.

Many display types do not provide simple and easy to understand measurements in line with gamma theory. One primary issue concerns how gamma is commonly calculated relative to white. Plasma, CRT, variable-backlight, and adjusting-iris displays can all potentially vary the brightness of white depending on the image displayed, which brings into question the validity of the reported gamma information when the measurement pattern introduces brightness changes. If the brightness of white changes, then generally the target values for all the other video levels also change when intending to calibrate to a certain gamma profile (say 2.3 display gamma). My original post generally focuses on how I view the current situation for gamma calibration, where people commonly attempt to calibrate various displays that vary the brightness of white depending on the image displayed using patterns that can introduce brightness changes.


There will always be some minor expected variation in even the best measurements, and I consider reported gamma values varying by 0.02 as very reasonable and unimportant. Based on the way I look at the gamma graphs from Chad B and Zoyd, their information seems to suggest a significantly larger amount of on screen variation during operation than 0.02 gamma. Personally I suspect the gamma variation shown in Chad B's and Zoyd's measurements is enough to be visible when watching the display, and if their data was put in the sort of terms that people use when discussing error, it would likely be significant. I'm not intending to bash plasma, because if problems from other display types, like non-uniformity for fixed-backlight or fixed-iris displays, were put into terms of error such issues would likely also fall into the significant category, but those issues simply don't affect typical measurements in the same way observed with plasma and other displays that vary brightness depending on the displayed image. Realistically I doubt if the gamma variation shown in Chad B and Zoyd's measurements can be corrected by most end-users, but also I don't see trends in Chad B's data to suggest that windows or APL-only patterns necessarily result in the best available compromise for how plasmas may operate. Since plasma gamma variation is likely easier to measure than some other display performance issues, personally I don't see the harm of looking further into if there happen to be ways to arrive at generally comparable gamma measurement information across various displays.
Edited by alluringreality - 3/16/13 at 5:48pm

The original post goes over my reasoning for the statement, start at "By this point I don’t know if anyone is still following along..."


My comments primarily focus on patterns where what you see on screen is what you measure, and I have not focused on typical window patterns in my comments. Window patterns primarily contain two video levels, where usually there is one level in the center for measurements and black for the surround. Generally windows vary average picture level (APL or average of the video information) and average luminance (AL or average light at the screen) by the size of the pattern. For example a 1% area window varies APL & AL by 1% between a pattern for black and a pattern for white, and a 15% area window will vary APL & AL by 15% between black and white patterns. A variation of only 1% in APL & AL may prove insignificant on certain displays, but my opinion is that something like a 15% change in APL & AL is rather crazy in relation to how usual video content displays.


APL stands for average picture level, which just means an average of the video information. Personally I think only focusing on APL may be misleading in some circumstances, so I like to at least check what is going on after the non-linear video information is converted to linear light at an ideal display. I have been using AL as an abbreviation for average luminance in this post. I'll suggest that the measurement data from Chad B does not directly support the idea that "the display will react accordingly depending on the APL of the image". Read the original post and then return to the following comments on the GCD 1% APL gamma graph from Chad B's measurements.

The GCD 1% APL appears to use a 1% area window on a constant background, which seems a similar concept to the patterns that Zoyd was measuring. The GCD 1% APL is around a 25% APL and 5%-6% AL (2.2 gamma approximation). If the display reacted according to APL, I would expect the gamma graph to look similar to something between Chad B’s APL and the AVS L APL. Instead the shape of the GCD 1% APL best resembles dark patterns like the DB Ramp Dark or 2% Window, and it also shares the high Y Max fL like the dark patterns. If the GCD 1% APL was averaged to a single gamma value like Zoyd was doing, it would probably come nearest to the AVS S APL, which is what the AL number might suggest. Personally I cannot say that either APL or AL fully describe how the display is operating in the measurements from Chad B, and even Zoyd seems to agree that some video content likely trends toward brighter overall images than represented by the GCD 1% APL. It's entirely possible that the The GCD 1% APL may be a reasonable approximation for some video content. Like the other patterns I discussed in the original post, the GCD 1% APL also appears above the horizontal reference line, which may suggest that plasma and fixed-backlight LCD measured to identical values using the 21% APL pattern might result in the plasma actually having a generally higher on screen gamma than the LCD when viewed side-by-side.


I agree the document is using a 1% area square for the gamma measurement pattern on 16:9.


Actually both graphs under "2.3.2 Presentation of the Measurement results" have reference lines for 2.2 and 2.5 gamma (spec. value = 2.35 +/- 0.15).


Personally I consider large windows likely worthless on a majority of displays, but it is the historical way of doing things, which is part of why I suspect typical gamma calibration attempts likely result in poor correlation on many displays.
Edited by alluringreality - 3/17/13 at 12:00pm

I'll suggest some LCDs or projectors offer settings to use fixed or variable lighting. Depending on the selected setting, they may operate more in line with either category 1 or 2.


I agree there are probably better ways to look at similar information, and error forumlas are based around Lab. Since I don't have a plasma to measure myself, the information from Chad B and Zoyd seems the best available documentation. Because Chad B’s APL and AVS L APL measurements have a similar Y Max fL numbers it's easy to conclude the on screen image has observably changed without calculating error. Zoyd's measurements appear to support the possability that on screen gamma changes on his plasma depending on the image displayed, since his graph is not approximately horizontal lines and each of his measurements are nearly fixed (APL & AL varies up to 1% during the measurement run).


Actually typical gamma graphs simply represent relative Y measurement data, and usually the information is related to a single white measurement. It would be possible to create multiple measurement patterns that contained a background, a gray level, and white. Instead of creating a typical gamma graph by comparing multiple gray levels to a single white, you could compare multiple gray levels to corresponding multiple white measurements. I will agree that more specialized measurements might generally represent a display's Electro-Optical Transfer Function, but I consider the measurements from Chad B as reasonable support for the idea that a gamma graph does not necessarily represent a display's EOTF similar to how it will tend to function with most video content.
Edited by alluringreality - 3/18/13 at 9:01am

The common measurement procedure obtains a gamma value using two images. Commonly people measure a gray video level from one image and compare it to a white video level from a second image, and the images can be entirely dissimilar to how the display will operate with the same video levels on typical video content. Everyone seems to agree the procedure results in erroneous data using fields on plasma, yet people do not appear to question how following the same method likely introduces similar error on a smaller scale with other measurement patterns.

Maybe the patterns from Spears and Munsil will meet their claims. Like I said in the last reply, a way to test the idea would be to embed a common size video level in the image and measure how much it changes during the entire measurement series. A common 1% area white included on a series of fields can be expected to vary significantly on plasma, depending on which gray image from the measurement series is displayed. Other dissimilar images will also vary to some extent during the measurement series, and the main issue I'm asking is if the variation is significant. I tend to think some of Chad B's measurements would be considered significantly different when compared using error values.