Metering tonality...18% or 12%?

+1 to Wayne. I was just starting to think about how I would say the same thing when I got to his post.

The matter is further complicated when we talk about color images because it is influenced by gamut size and choice of primaries and white point. 18% grey is 119 in sRGB and 117 in Adobe RGB. A further complication is that the sRGB TRC is not perfectly gamma 2.2, it is linear at the bottom and rolls off to roughly 2.4 at the shoulder, but averages around 2.2. And ProPhoto RGB is gamma 1.8 and D50, which puts 18% grey somewhere down in the 90's. BTW, LR's histogram is Prophoto with the sRGB TRC - as told to me by Andrew Rodney - which puts medium grey somewhere around 103 (IIRC) and is one of the major reasons for the LR team's decision to make RGB values available only as percentages, rather than hard 8 bit numbers.

Wilt wrote in post #12081168
OK, Wayne, got your discertation about Gamma and encoding. But explain, please, for further understanding:

• if there are 256 levels of tonality within the scale of absolute black to absolute white, 128-128-128 in pixel color encoding is the midtone that is like 18% gray.

Maybe you didn't get it yet. Your sentence is absolutely false. On this fictional absolute scale (the sensor is linear, but linear is real only in analog, and in RAW data at the sensor), 18% is NOT 50% of anything. Linear 18% is 18%.

18% of 255 is 46, not 128.

One stop is 50% of the light. 18% is closer to three stops down... which in linear data, this 18% is at 46, which is shown by 255... 128...64...32... (in linear data, which we can never see).

We can only see a histogram after it is converted to RGB, and then gamma encoded, and (in a perfect world), 18% is shifted to 117, which is coincidentally 46% of 255.

And this so-called magic spot of 128 has been shifted to about 187 anyway. All as previously explained.

What is wrong is that 18% is not 50% of anything. It is 18%. And our viewing scale is greatly distorted by gamma encoding.... which shifts it to 46%, for reasons having absolutely nothing to do with middle of anything. The "middle" incidentally is shifted to be around 187.

Try this. Carefully expose some white subject to just align the data at 255 on the histogram. Then underexpose it exactly one stop (50%). The 255 end does NOT shift to anywhere near 128. It goes to around 187, or around 73%. See? (the 187 cannot be a precise number due to all the other factors working in there).


The gray card is an antique analog contrivance for ink printers. Speaking of absolutely, there is absolutely nothing digital about a gray card. It was said to be the color of reflected ink that the logarithmic human eyes and brain perceived as "middle" gray (but the data is still 18%). It was popularized for photography by Ansel Adams in the 30s. No better tools existed at that time. The only way the data could be viewed was to print it on paper, and look at it with human eyes. Nor did digital histograms exist.

But someone used the word "middle" regarding this gray tone.

When digital histograms were invented, someone said middle about their center point.

Then everyone just assumes every use and meaning of the word middle naturally always is related to mean the same thing. It was simply poor understanding of the crudest kind... There is absolutely nothing middle about 18%.

Wayne, what explains the midtone (18% step, similar to the midtone background area) having R-G-B 128-128-128, while black is 0-0-0 and white is 255-255-255?

That gray scale is (let's say) a contrived imagination product. It existed decades before digital. Try photographing that gray scale chart. You can of course duplicate the analog chart in an analog printed copy on paper - BUT - then examine your histogram. You will NOT, repeat NOT, see equally spaced lines like you imagine. Your analog chart tries to show what linear data might look like (for our imagination), but our histogram is A) digital, and B) gamma encoded.

Repeat: (takes maybe only 5 minutes)

Try this. Carefully expose some white subject to just align the data at 255 on the histogram. Then underexpose it exactly one stop (50%). The 255 end does NOT shift to anywhere near 128. It goes to around 187, or around 73%. See?

Seems this real result really ought to be factored into any reasonable explanations of how the real world works.

WayneF wrote in post #12083268
Repeat:

Try this. Carefully expose some white subject to just align the data at 255 on the histogram. Then underexpose it exactly one stop (50%). The 255 end does NOT shift to anywhere near 128. It goes to around 187, or around 73%. See?

Seems this real result really ought to be factored into any reasonable explanations of how the real world works.

But Wayne, we are not talking about an exposure system which is only 2EV wide in dynamic range, so -1EV down from max white would not result in the midtone color gray. If we had a 2EV dynamic range system, I would be in full agreement with you about the underexposed -1EV result, but we are not (even for the cheapest of cameras)

In the Zone System middle gray was the midpoint Zone V, of what Ansel Adams himself described as a geometric scale from black to white. Your Gamma and that scale is recognition of the geometric nature, but it does not change our perception of middle gray or where that happens to end up in the 0-to-255 8-bit scale.

Wilt wrote in post #12083297
But Wayne, we are not talking about an exposure system which is only 2EV wide in dynamic range, so -1EV down from max white would not result in the midtone color gray. If we had a 2EV dynamic range system, I would be in full agreement with you about the underexposed -1EV result, but we are not (even for the cheapest of cameras)

In the Zone System middle gray was the midpoint Zone V, of what Ansel Adams himself described as a geometric scale from black to white. Your Gamma and that scale is recognition of the geometric nature, but it does not change our perception of middle gray or where that happens to end up in the 0-to-255 8-bit scale.

One of us is not hearing anything from the other. I hope that is not me. I know the wrong poor explanations have been around for decades. It never mattered until we had digital and histograms. But now it matters. At least, there is a big distinction.

You insist on hanging on to your notion that 128 is middle gray.
And also insist that 18% is middle gray.
Come on now, which is it?
(answer: in a histogram, neither).

My understanding is that 1 stop is exactly 50% of the light. By definition, that 50% is midpoint. I doubt you disagree with that.

So again, simply carefully expose white to be at 255, and then underexpose exactly one stop. How do you explain what you see happen?

My way is simple. In analog data, and in the linear RAW sensor data, one stop down is 50%, and in linear digital, it is at 128 (half of 255). Just like we always imagined.

The problem is that we can never see that. We only see RGB data that is gamma encoded, and 18% at 46 goes to 117, and midpoint at 128 goes to about 187.

So in any discussion about what the histogram ought to show,
we simply must understand what it is that it does show.
It only shows gamma encoded numbers.
You seem totally unable to accept that.

WayneF wrote in post #12083268
That gray scale is (let's say) a contrived imagination product.

Ahem. As the creator of that gray scale, I can assure you all that it was indeed contrived. The primary purpose of it is to allow the viewer to adjust a monitor by viewing the a, b, y and z bars.

I chose to call the background and the middle bar "middle gray" to avoid confusing people. I have no idea what the reflectance of it would be when printed, but it's not 18%. Its shade is RGB 127, 127, 127.

-js

Wayne,
I think that your explanation of Gamma and mapping is one thing, the mid-tone reprensentation with an arbitrary 8-bit representation of 256 tones is a different concept. The two coexist.

My point is that the two concepts need to coexist. The 8-bit 256 levels simply maps a variable range of dynamic capture (depending upon the capabilities of the sensor and its supporting circuitry) and allows all monitors or printers to portray the two extremes of white and black, and the 254 other divisions in between.

I think that Tony's question about 'how much below' is what resulted in this dual concept apparent conflict, because he was trying to equate a tonality difference to an EV difference...yet that would vary depending upon the specific dynamic range capability of the sensor and circuits.

'18%' reflectance is the value that came about thru use of sensitometers, the quantification of the mid gray tone in between 100% reflectance and 0% reflectance. In that sense it is not arbitrary, it was the result of quanfication using instrumentation. It was the middle zone in the ten zone Zone System. Our digital cameras cannot equal those ten zones readily, particularly unless RAW capture is used. Whether we have an eight zone system (digital), or a nine zone system, or a ten zone system (B&W film), the 18% reflectance (as measured by a reflectance sensitometry) is the mid tone value.

And in digital representation, 1-256 has mid value of 128, and 0-255 has mid value of 127. Whether 127 or 128, macht nicht...the mid point of a numerical sequence.

Wilt wrote in post #12084576
My point is that the two concepts need to coexist.

Of course both concepts do exist. One of them is just so wrong-headed. I have shown you how to do the simplest test to show the obvious truth of that. You don't want to comment on it.

You still insist that the center of a histogram is both 18% and 50%, at the same time? (when in fact, gamma prohibits both notions). You don't want to comment on it.

I suppose you must deny that our RGB data is gamma encoded. Standards of sRGB with gamma 2.2 simply must not apply to your images? You don't want to hear about no stinkin' gamma. You don't want to comment on it.

Pity... but I cannot get you focused on the actual indisputable facts. You are stuck with your graycards, grayscale, Zone System, etc, frozen back in the analog days. There is no concept of a histogram in analog data. Digital RGB data (which definitely is gamma encoded, I promise) and associated histograms showing pixel counts from that data, are a different system, with its own rules now.

Ultimately, our picture at monitor or printer does necessarily convert back to decoded analog for viewing with the human eye, when 18% possibly may have some significance (it has no other significance). Digital does try to reproduce the analog image exactly, but while stored as digital, there are a few different concepts, new stuff which you should learn now.

Good luck with that. You will need the concept of gamma to understand any (visible) histogram values. No way around it.

Wayne,

What is hard to grasp about using a reflected light sensitometer to measure black as 0% reflectivity, measure white as 100% reflectivity, and the middle tone as having 18% reflectivity? It is no different than a measuring cup with 0 fl.oz. or full at 8 fl.oz, and half full at 4 fl.oz., is it? And such sensitometry predated digital imaging, being used for film for decades (which had their own gamma values!)

I am denying nothing about gamma encoding, which is why I had not commented on that point. (not to mention that I do not have enough knowledge in this area to deny anything on that topic!)

Let's set this discussion aside, as it is apparent that there is no direct correlation between 1EV of exposure change and a 26 step or 256 step or 4096 step tonal scale between the extremes of black and white in continuous grayscale representation.

As I stated at the beginning of the exchange with you, I was seeking understanding, not a debate.

Wilt wrote in post #12085366
What is hard to grasp about using a reflected light sensitometer to measure black as 0% reflectivity, measure white as 100% reflectivity, and the middle tone as having 18% reflectivity? It is no different than a measuring cup with 0 fl.oz. or full at 8 fl.oz, and half full at 4 fl.oz., is it? And such sensitometry predated digital imaging, being used for film for decades (which had their own gamma values!)

The difference is that while you might measure the 18% card at 18%, but then it will not measure 50%, no matter what name you call it. Yet you want to call it "middle gray", and assume it somehow must be at 50%. It is not, it is 18%. The word middle has many meanings.

Film gamma is a concept entirely different.. is the slope of the liner film response curve, a measure of contrast. Kodachrome had more contrast than Tri-X. This is not remotely related to encoding digital data to be different numbers. (many different curves are called gamma, it just just a greek letter... just like X in algegra).

See http://en.wikipedia.or​g/wiki/Gamma_correctio​n about gamma.

It is really not a hard subject, yet it always seems to come out as being difficult. The big boys seem to complicate it.

Trying to be helpful - a very short simple version:

Earliest work knew that CRT tubes were not linear. Only the strong (brightest) signals showed up on the face of the CRT, and dim values simply disappeared. This was not so bad for oscilloscopes, which typically only showed one signal level, but grayscale involved many different tones, over a wide range. CRT was unacceptable use for grayscale. Of course, there was no grayscale until early television.

Earliest television solved this by encoding the transmitted signal oppositely. They boosted the weakest values more and the strong values less. This boost curve was called gamma. Each data value was increased using an exponent of roughly 1/2.2. Exponent is near 1/2, which is close to concept of using square root of all data values. Technically strong signals were reduced more than weak, but in effect, weak values were boosted more than strong signals.

-1 stop = (0.5 ^ 1/2.2) = 0.73, x255 = 187, 73% of 255 full scale
-2 stops = (0.25 ^ 1/2.2) = 0.54, x255 = = 137, 54% of full scale
18% Gray card = (0.18 ^ 1/2.2) = 0.46, x255 = 117, 46% of full scale

That 1/2.2 gamma curve boost corresponded to about how much low level signal the CRT tube lost, and it came out looking just about right on the screen of the CRT. Doing this encoding in the transmitter once was much better than adding circuits in EVERY television receiver to do it. Boosting the low level signals helped noise on the transmitted analog signals too. The data was gamma encoded. Then all a CRT had to do was to display it, and the losses were precomputed, and it came out right.

Computer CRT monitors had the same losses, and when we started getting into computer images (popular around 1990), this gamma became the world standard for all images... All analog TV signals, and all computer digital data was gamma encoded, with exponent of roughly 1/2.2. This compensates for the CRT losses which were roughly exponential with a 2.2 exponent.

This means EVERY RGB image has been gamma encoded. The encoded data values are NOT the same as the linear data values. The histogram shows different numbers now.

Todays LCD are linear... no losses, they do not not need gamma. However, all of the RGB data in the world is gamma encoded now, digital TV signals too, so the LCD has to deal with it too. A LCD monitor specifically must decode gamma first, where a CRT does the same by merely showing it.

Perhaps the day might come that we don't use gamma, but I doubt it. LCD is pretty recent, and they have to deal with the data that exists. And printers need some degree of it too, may be not as much, but most of it. They are programmed to deal with gamma 2.2 data, and to adjust it to their needs. Easier to continue than to change every thing, and I mean everything.

Long story short... Analog data (except television) has no concept of this gamma. The camera sensor is linear, and it has no concept of this gamma (in the RAW data at the sensor). The word linear has two meanings in video, one is the "graph as a straight line" idea in math, but mostly in video, it also imples "not yet gamma encoded".

But all subsequent uses in the world do, and all RGB data is gamma encoded. All cameras and all scanners output gamma RGB images. All RGB images are gamma encoded. All histograms show gamma encoded data. The histogram numbers are simply different than the linear numbers we might imagine. Just how life is.

It will be a good thing to believe it too.

If you re-read all of my posts in this thread, as I never said 18% reflectance was 50%, I merely said 18% reflectance was in the middle of a 10 zone Zone System scale and also in between absolute black and absolute white.
Ansel Adams called it 'middle gray' and assigned that to Zone V, and I am merely using commonly accepted usage of the definition per classic photography film and paper print definition.

As for the history of gamma encoding, I do not dispute your statements.

So please, now apply the above concept to a system which permits 9EV of dynamic range...I see maybe 5EV in what you describe, and now want to understand what changes in the computation of a 9EV system.

Wilt wrote in post #12085926
Ansel Adams called it 'middle gray' and assigned that to Zone V, and I am merely using commonly accepted usage of the definition per classic photography film and paper print definition.

I believe that Ansel and his Zone System in the 30's predated digital, by several decades. My guess is that he never saw a histogram. Things change.

Nothing I have said relates to EV of dynamic range in any way. I am just telling you the numbers you imagine to be in the histogram are different numbers, and the simplest one stop test underexposure of 255 will easy show that to be true.

In the simplest way to look at it, 8 bits linear can show 8 EV of range (8 powers of 2), but that does not mean the contained data has 8 EV range, or maybe the original data may have more range than can be shown. In the same way, just because you subscribe to Ansel's Zone System does not automatically give you ten stops.

EV is NOT my argument, and I do not care to pursue it. I can say that Charles Poynton is a huge name in gamma, his name is always quoted as reference. He is typically a hard read, you really have to pay attention to the precise meaning of every little word, and might get what he meant the tenth time.

His claim is that non-encoded data needs 14 bits to show change from black to white, but that 8 bits gamma encoded can show it. He shows his reasoning, and I leave it up to you and him,
see http://www.poynton.com​/PDFs/GammaFAQ.pdf
item 13 on page 7.

I imagine this was sort of true in the days of CRT monitors.
However LCD monitors must decode it into 8 bits first, and would seem to lose something at that point. Also, many/most LCD monitors today are only 6 bits internally.

This aspect is not my fight. You are on your own.

I'm trying to wrap my head this stuff, but I don't have the background that you guys have.

Something that seems evident from what Wayne is saying is that we are not dealing with linear data here and so shouldn't try to "jam" linear terms into it.

So, as we know, the histogram is not a linear graph, and, for example, a gray object/card/whatever that spikes in the dead center of the histogram is not linearly half way between black and white but only shows there due to the gamma correction. Linearly, one stop lower than white is half the light, and so a linear scale would show that as "half way" to black.

Wayne, am I on the right track here?

So Wilt started all this with explaining the difference between the 12% calibration and the 18% gray card, and I see that this is happening in the digital non-linear realm -- the idea being that the 1% gray card with the proper calibration would land in the middle of a histogram. But, of course, this doesn't represent any linear representation but the gamma-corrected digital histogram. As we know, a stop down from white won't land in the middle but a fair amount to the right of middle.

So, what actually defines an 18% gray card seems to be lost in space between linear and non-linear. Is it correct that on the linear scale it is not one stop less light? It sounds like it. But how does that fit in the approach that to get from say dead center on the meter/histogram up by one "notch", we double the light collected, one stop, and we land halfway between white and black? Does this mean that 18% gray is best defined as two stops less than white, as opposed to "medium", and yet we explain that it should show up as dead-center in the histogram because the histogram is non-linear, and it should show up in the RGB color spaces as "half" of white because they are non-linear because in fact it is not half of white?

Whew! Did what I say make any sense at all?

Your confusion, Tony, seems to be where this thread has landed most of us.