Histograms

[Tim_Lookingbill]

The reason the brightest first stop doesn't have the most tones is that the image doesn't

have any bright=near 255 data.

 

And besides a luminosity channel histogram is a relative representation of the composite

of all three RGB data channels before a gamma or brightening curve is applied.

 

If you do a search entering Histogram in Photoshop's offline Help, click on "Checking scan

quality and tonal range". It'll explain Photoshop's way of representing captured data in its

histogram. My understanding from this is the histogram is a relative representation

between how Photoshop represents image data compared with how it's represented by the

linear distribution grayramp graphic shown on page 2 of Bruce Fraser's pdf. These two

representations of image data distribution don't necessarily have to line up exactly

between each other due to the variances of any given scene.

 

�I think what's more important to notice about Bernie's histograms is how different the

change in appearance between the two histograms are just by adjusting by one f-stop. If

you were to measure the width of the two main peaks at the same points with a ruler it's

pretty close to halving or doubling in width how ever you want to reference the two.

[bernardwest]

Van, I've already mentioned that I'm not versed in the zone system. I really don't see what Zones have to do with it. But even not knowing much about zones, I can see that you are applying them wrongly to the digital realm. Zones (i believe) are an arbitrarly defined system (ie. the image tones are arbitrarily divided into some number of zones. Is that right?). There is, however, nothing arbitrary with the histogram. The histogram DOES fully represent EVERYTHING that goes on in a digital capture (linear raw). You don't need to know ANYTHING else (except for one thing which I will mention shortly). Where you're going wrong with Zones is highlighted by your quote of the 'Nikon D200 vs Tranny film' guy. I haven't read the link, but from his quote it is clear that he is talking about a rendered raw file (that is, wb and gamma curve applied). The rest of us aren't talking about rendered jpg's or tiffs. We are talking about pre-rendered linear raw data.<br><br>

 

<i>"The histogram gives us no idea of the shape of the curve and more importantly gives no clue to shoulder/toe performance?.. this is why were having problems. This is the first step Reichman needs to do (set up a chart) to clarify".</i><br><br>

 

There is no need to set up a chart, because it is really very very simple. The sensor is a LINEAR measuring device. ie. The chart is a STRAIGHT diagnal line, with NO shoulder or toe. Double the input and you double the output. Halve the input and you halve the output (with some little adjustment for black point clipping). This is why if you stop down 1 stop (halve the input), you halve the number of levels that are recorded (in relation to the original exposure).<br><br>

 

I mentioned earlier that I was starting to doubt myself, but it's alright, I've come good. Sometimes I get so carried away in thought experiments that I forget that I can just go and perform an actual physical test. Which is what I did, and which I recommend you do to Van. The only problem is that you really need to analyse the raw data itself, not some rendered file (even if it is only a linear tiff, like I did). I'm going to download a raw analyser and confirm this once and for all. But just using my linear tiff's it is clear that noise in the overexposed (then underexposed in raw converter) image was much less than the 1 stop less exposure.

[bernardwest]

Also Tim, if you look at the mean and median values of the two histograms you can see that the more exposed one is in the region of 2x the less exposed one. The fact that these don't match up to exactly two times is 1 - like i said, these are tiffs, so have had wb performed and will vary a bit due to this, and 2 - like you said Tim, the luminosity is some sort of average. Ideally this sort of test will be conducted on pre-rendered raw data (which is what I intend to do soon).

[bernardwest]

Van,<br><br>

 

<i>Also when looking at the histograms you provided, it DOES NOT show a 1 stop change (you made a mistake somewhere).</i><br><br>

 

It DOES show a 1 stop change, because the exposures were 1 stop apart. No mistake. I can post the original images with EXIF if you want.

[van_camper]

Hi Bernie, you mentioned that you do not know much about zones or see where it is important. However, everyone I know uses it all the time to place a highlight on zone 7 approximately and let the values fall (just as important for slides as it is for dslr). Your lack of interest in the characteristic curve and histograms is apparent. Your belief that there is a 1 stop difference between the tonal values 128 and 255 on a histogram are incorrect (as I said there are approx 32 tonal values between each zone or f/stop?so a 4 stop difference between 128 and 255). I also mentioned your two histograms do not support that that they were shot 1stop apart (impossible-an error somewhere). I do not see pursuing this any longer, because I do not want to back up and start again and because I can see you have no desire to intensely get involved in curve theory (histograms is not enough). In photoshop, when you look under CURVES, that didn't just pop up from know where. It is worth understanding. It is an involved subject.

 

"This is why if you stop down 1 stop (halve the input), you halve the number of levels that are recorded (in relation to the original exposure)." Not correct, you half the brightness on the x-axis with each stop, which shows the corresponding change on the y axis (tonal values). It doesn't half the number of tones, rather it changes the value of the tone (eg- zone 7 becomes zone 6). This is why your getting confused, your ignoring the study of sensitometry and think you can understand how things work without that knowledge. You want to do it all with a histogram which is incomplete although you believe otherwise).Film and digital has not changed, and digital is getting closer all the time to imitate film?.that is why our handheld lightmeters still work, ISO values match film, people use dslr for polaroids and film exposure.

 

"Where you're going wrong with Zones is highlighted by your quote of the 'Nikon D200 vs Tranny film' guy. I haven't read the link, but from his quote it is clear that he is talking about a rendered raw file (that is, wb and gamma curve applied). The rest of us aren't talking about rendered jpg's or tiffs. We are talking about pre-rendered linear raw data."

 

Well, unfortunately you haven't read it. He summarized to say slide and digital files are similar (except for small differences he found in toe/shoulder of curve). I find your having a hard time understanding RAW and JPEG conversion. Leaving out compression with jpegs, all that has happened is gamma was applied. Gamma means slope change (which I can tell means nothing to you unless you have read up more on sensitometry! I understand RAW perfectly, work in it all the time, but also in large format film. I discussed it before and you ignored my explanation.

 

I'm outta here, Reichman never showed the courtesy to spell it all out on paper to prove himself. I've already spent way more time then he did.

 

Good luck with your shooting. I enjoyed the discussion.

[bernardwest]

Van, you're right in one thing. That is, we've beaten to death this discussion!<br><br>

 

However, I do understand curves, although digital better than film (I do understand that film has a toe and shoulder, but MORE importantly that a digital sensor has NO! toe and shoulder - a point which you seemed to not address in your last post... are you trying to ignore this reality??). On that last point, digital will have a shoulder at least, but only AFTER it is rendered and the gamma tone curve applied (ie. a jpeg et al). This is the point you are missing. ETTR doesn't apply to jpegs. It applies to linear raw data.<br><br>

 

You need to come out and state definitively whether you believe a digital sensor records information in a linear manner. If you do, then you can no longer continue to deny my statement "This is why if you stop down 1 stop (halve the input), you halve the number of levels that are recorded (in relation to the original exposure).". If the sensor is linear (which it is) then it can function in no other way. In a linear graph of X vs Y, input vs output, half X equals half Y. No buts, no maybes. You can't possibly argue against this. Do you believe the digital sensor is a linear device?<br><br>

 

<i>I find your having a hard time understanding RAW and JPEG conversion. Leaving out compression with jpegs, all that has happened is gamma was applied.</i><br><br>

 

Do you understand gamma correction? In my linear tiffs I posted before, one had a maximum level of 195, and the other had a maximum value of 101. Do you understand what happens now? Both 101 and 195 get mapped to higher values, but NOT in a linear fashion (ie. the proportional AND absolute values of increase will not be the same). Therefore, when you are comparing exposure values between two jpegs, you aren't comparing apples to apples. That's why this argument isn't valid for jpegs, and why you should stop refering to them.<br><br>

 

Finally, in relation to Zones, I stand by my original statement, that being that they are arbitrary and can't be used in a mathmatical argument. To get a bit of background on them i looked them up in Wiki (I know, hardly authorative, but close enough), and this confirmed what I previously knew about them. 10 Zones from darks to lights. How this relates to digital I don't know. It certainly doesn't seem to relate to your specification of 32 levels in each zone. And on your specification, 256/32 equals 8 zones/stops. This is imprecise science if ever I have seen it. Are you saying that there is 8 stops of dynamic range recorded by every camera ever, regardless of make or model. A quick search of the internet (not even that wide - just try dpreview and luminous-landscape) will show that different cameras record different dynamic ranges. I really think you need to forget about zones and stick to mathmatically precise digital concepts.<br><br>

 

Having said all that, I have enjoyed our discussion (Tim's too), and it has forced me to really think about how things work and investigate a couple of new things too. I will post the tiffs with exif so you can see for yourself. Keep in mind they aren't true linear raw as white balance and colour space conversion has occured. Now if you don't believe the files I post, do your own test and you will see there hasn't been any mistake.<br><br>

 

cheers, Bernie

[bernardwest]

heres the 1 stop over.

[Tim_Lookingbill]

Van,

 

To put it simply based strictly on my own unscientific observation, it's not about the

behavior of film, zones, histograms, curves or digital data, it's more about the behavior of

electronics IMO.

 

Film is developed through a light sensitive chemical process. So it's pretty hard to make

the connection that silicon chips, Bayer filtering and voltage signals converted into pixel

data would behave the same exact way.

[van_camper]

"Film is developed through a light sensitive chemical process. So it's pretty hard to make the connection that silicon chips, Bayer filtering and voltage signals converted into pixel data would behave the same exact way."

 

Well, pull out your handheld light meter, try it out. Does it still work with digital? Is there not a relationship (1 stop increases perform equally well in digital, and zone system is fully being used)? The equipment is different, but the sensitometric curves are following the film curve as close as it can with todays technology (I did not say EXACT). The Fuji S5 with two sizes of sensors has made it even closer. If you do not see a similarity, why do we have curves in PS showing the same relationship (less shoulder and toe)? The real problem is your caught up thinking histograms and have not even considered discussing curves straight from photoshop. Why is this ignored, and what do you think I was talking about? I have discussed digital behavior all along (but bringing up similarity to film).The real problem is lack of knowledge in sensitometry, because issues I brought up, corrections I made were still fluffed off.When you continue getting the response that there is a 1 stop difference between value 255 and 128 on a histogram, then I know we have a problem. The value 255 and 128 is a one stop difference on the exposure unit scale, not on a histogram only where the horizontal axis is shown as 0-255 in increments 1,2,3,4... So good luck eveyone. Regardless of point of view, you still will get a fine image, and it was enjoyable. I recommend you read up on characteristic curves plotted by others for digital.

[van_camper]

Come on Bernie, please read what I write.

 

1."but MORE importantly that a digital sensor has NO! toe and shoulder - a point which you seemed to not address in your last post... are you trying to ignore this reality??). On that last point, digital will have a shoulder at least, but only AFTER it is rendered and the gamma tone curve applied (ie. a jpeg et al). This is the point you are missing."

 

Answer?please reread all articles above. I have said this at least a dozen times. I described the difference, stating it doesn't have shoulder/toe. But apparently you do not understand what I meant by the straight line portion of the film curve being the same as digital.

 

2. "Do you understand gamma correction? In my linear tiffs I posted before, one had a maximum level of 195, and the other had a maximum value of 101. Do you understand what happens now? Both 101 and 195 get mapped to higher values"

 

Answer?please reread what I said earlier.

 

3." Finally, in relation to Zones, I stand by my original statement, that being that they are arbitrary and can't be used in a mathmatical argument. To get a bit of background on them i looked them up in Wiki (I know, hardly authorative, but close enough), and this confirmed what I previously knew about them. 10 Zones from darks to lights. How this relates to digital I don't know. It certainly doesn't seem to relate to your specification of 32 levels in each zone. And on your specification, 256/32 equals 8 zones/stops. This is imprecise science if ever I have seen it. Are you saying that there is 8 stops of dynamic" .

 

Read the histogram scale, realize it is a tonal scale 1,2,3,4,5,6?255, not a exposure unit scale represented as 1,2,4,8,16,32?. Nuff said, been discussed before. I won't be discussing it for the 5the time. As for 10 zones and you not understanding how the zone system relates to digital, well, that takes the cake. I know it is time to stop the dabate. Lets see, every pro I know likes to place their highlight on approximately zone 7 (please note, I am referring to a digital camera). Where have you been? This makes it apparent that you do not know enough, do not read what I have said, and therefore I am wasting my time. I have even mentioned this same point earlier

 

 

By the way, just visually looking at your two jpegs, there is no way I can believe that the plants in both images are 1 stop apart. The difference in highlights between the 2 images is more then 1 stop apart(walls inc). If it were film, I would swear you expanded development. Nuff said.

 

END

[bernardwest]

Van, I'd love to keep debating this topic, but my wife says if I don't stop arguing with you I'll be spending christmas alone! Look forward to meeting in another ettr topic thread in the future sometime.

 

By the way, as you don't believe my results, do a test yourself, and post your results. I'll have a look, and when the wife isn't looking I might even add a comment...

[van_camper]

Hi Bernie, the comment about what your wife said was funny. She was right, and it was getting too involved...tiring for both of us. I didn't mean that I didn't believe you, just that looking at those two images (plant more specifically) seemed far more then 1 stop (and I've shot many over 35 yrs). Seemed more like an error somewhere along the way. I don't want to get into this any more, only way would be with beer as I said at the other forum, and a pencil and paper. We were frequently misunderstanding each other. Talking about sensitometry is like trying to explain calculus over the net...not going to happen it seems. More important, get some shots. Finishing up a nice one today (made up of a few bits and pieces to complete the composite)

 

Wishing you a Merry Xmas.

 

Van

[van_camper]

Hi Bernie, earlier you commented....

 

"It certainly doesn't seem to relate to your specification of 32 levels in each zone. And on your specification, 256/32 equals 8 zones/stops. This is imprecise science if ever I have seen it. Are you saying that there is 8 stops of dynamic range recorded by every camera ever, regardless of make or model. A quick search of the internet (not even that wide - just try dpreview and luminous-landscape) will show that different cameras record different dynamic ranges. I really think you need to forget about zones and stick to mathmatically precise digital concepts."

 

Refer to the following website for a better idea. He comments about 32 values between points on the histogram scale. Imprecise science, not really....that is the way it is. Notice 4 zones between value 128 and 255.

 

http://www.lashier.com/home.cfm?dir_cat=22078

[emre]

Van, that page is wrong. I hope you did not get your information from there. I count 4.5 steps from 0 to 128 (just take a ruler and see). How can nine steps of 32 add up to 256 values? Someone is making a mistake... Besides, I thought the Zone System had eleven steps (0-X), not nine?

[Tim_Lookingbill]

And also Lashier mentions he processed the nine step grayramp in sRGB which has an odd

ball 2.2 gamma shaped curve that when assigned to a regular gamma encoded image will

bump up densities by as much as 10RGB levels right around 5-80RGB gray.

 

Try it. Make a 21step grayramp in Photoshop AdobeRGB and assign sRGB2.1 and notice

the slight bump in the shadow area.

 

Whether converting to or assigning this space will induce inaccuracies within the math if

so referenced.

[van_camper]

"Van, that page is wrong. I hope you did not get your information from there. I count 4.5 steps from 0 to 128 (just take a ruler and see). How can nine steps of 32 add up to 256 values. Someone is making a mistake... Besides, I thought the Zone System had eleven steps (0-X), not nine? "

 

Hi Emre,you are right, I made a mistake. I was thinking zones 0-9, therefore thinking there is 8 intervals in between, which gives value of 32 (when you divide 255 by 8). It should have been zones 0-9, which is 10 steps, therefore I have 9 intervals inbetween giving a value of 28.3 between zones. Zone 10 is considered paper white, not a tonal value like 0-9 are. Ansel used zones 0-9 (The Negative, Basic Book 2, pg 14), Fred Picker used 0-9 (Zone VI Workshop, pg 10), and Phil Davis also 0-9 (Beyond the Zone System), but he prefers to consider zone 9 as paper white. I have seen others use zone 1-9, giving 4 zones above, 4 zones below. So long as you calibrate, it doesn't matter, it is not set in stone.

 

Regardless, we agree there are several zones between 128 and 255.

[bernardwest]

<i>Regardless, we agree there are several zones between 128 and 255.</i><br><br>

 

Maybe, but there is still only one stop difference between two differently exposed images. :)

[bernardwest]

Maybe I should rephrase what I just said. For two exposures of an identical scene, if one displays a certain highlight at 255 and the other shows the same highlight at 128, then the exposures will be 1 stop apart.

[van_camper]

"For two exposures of an identical scene, if one displays a certain highlight at 255 and the other shows the same highlight at 128, then the exposures will be 1 stop apart."

 

Bernie, on a Exposure Unit scale (1,2,4,8....128,256,512) you would be correct and there is a 1 stop difference between the values 128 and 255(see chart I included above).However, the histograms horizontal axis shows a "tonal" scale 1,2,3,4,5,6,7...255, not an "exposure" scale.

[bernardwest]

For images captured by a linear sensor 'tones' and 'exposure' are equivalent. The most exposed thing on a linear sensor will have a tonal value of 255. Something with half the exposure of full saturation will have a tonal value of 127.

 

Do the test yourself, and you will see I am right.

[van_camper]

"For images captured by a linear sensor 'tones' and 'exposure' are equivalent"

 

Wrong. Exposure and tones are seperate variables. Check your "curves" in Photoshop. The horizontal axis is the exposure unit(brightness) scale (1,2,4,8...2x factor). The vertical axis is density, referring to tones (not 2x factor). Unfortunately they do not label it. Thus curves shows the relationship between exposure and tonality. This I have mentioned several times.

 

"The most exposed thing on a linear sensor will have a tonal value of 255. Something with half the exposure of full saturation will have a tonal value of 127. Do the test yourself, and you will see I am right."

 

First, to be correct, replace the word "tonal" with "exposure", they are different scales. Next,why test? Look at the value 128 in the middle on the tonal scale (histogram), and you can see several other tones above 128 that represent tonal values for zone 6,7,etc (each 1 stop difference). Better yet, load an image with lots of white in it, start with 255 and move slider slowly, you will see before you reach value 128 various tones representing zones 6,7,8 while looking at the white area on the monitor. Do you really believe a 1 stop decrease (moving from zone 9 to 8) looks like value 128 on the histogram(which is middle grey, zone 5). The histogram scale is arithmatic....1,2,3,4,5,6 (the 2x factor does not apply the way your abusing it).

 

The real problem is too many are trying to discuss issues about sensitometry without understanding sensitometry. Without this, the relationships between RAW, JPEG, etc are also confusing.Worse yet, Reichman has misinterpreted the use of the "exposure unit" scale (which is what I was originally disussing).

[van_camper]

By the way, HAPPY NEW YEARS to everyone!

[bernardwest]

I'm not sure which curves you are looking at. The curves in my version of photoshop have 'input' and 'output' tones for its axes.

 

The only thing that will resolve this is for you to test it. Take a shot of anything that doesn't clip, reduce exposure by 1 stop, and take another shot and investigate the linear raw data. You will see the most saturated tonal value will be reduced by roughly half.

[van_camper]

"I'm not sure which curves you are looking at. The curves in my version of photoshop have 'input' and 'output' tones for its axes. "

 

So does mine. Those are just intersection coordinates (not describing both axis), which tell me you are not understanding the relationships between the two axis I had explained earlier. When you tell me that tones and exposure are equivalent it only supports my belief. What happens if contrast is increased? Then the 1:1 relationship no longer exists. This is why there are two variables describing curves. The input of one causes an effect on the other, but the output value can be higher or lower. If the input value is lower then the output value, then it is describing a curve with increased contrast (and vica versa).

 

Unfortunately I noticed were having continued differences with terminology, and sensitometry (differences in comprehension of zone system, exposure unit scale, etc). I do not think were going to satisfactorily come to an agreement. I'm sure with a pen and paper (and beer of course), we could satisfy our differences, but not here.

[tom_t4]

Seems as soon as someone mentions a histogram, the topic of ETTR comes up. I see Bernie and Van have been going at it some time, and although I have not read everything I would like to comment.

 

The first thing that caught my interest was Vans comment about Reichman not using the scale appropriately in building his chart. I would have to agree with Van. An exposure scale has been substituted for a tonal scale, they cannot be both.One refers to quantity of light required for a 1 stop increase (doubles), and the other refers to the number of tones remaining after a 1 stop change which doesn't even make sense. Worse, the sale canot mean both. Also exposure for each pixel is random, not sequential as Reichman describes in his declining table. Each pixel may receive the least or the most exposure depending on the scene. Neither the RAW or Jpeg histograms support his theory. The right side of the histogram should show the highest peaks if you believe in ETTR, yet they do not. Since his theory employs a pattern (a descending order of data), I would expect this same pattern showing itself in the histogram (a record of our exposure) which it does not.

 

Regardless, whether you believe or not doesn't really matter.Ask yourself how often can you ETTR. You can shoot all winter and not get the opportunity in landscape photography. Perhaps you can employ it 30% of the time, but does that mean the other 70% of the exposures are of lesser quality? ETTR hs been around a few years now, yet before that we still had digital backs and ETTR was unheard of.The mfr installed the meter for a purpose. With multi zone metering,and all the other bells and whistles, I am sure ETTR would have been added if it had real value. Do you believe the mfr or Reichman?