Histogram: Luminosity and Color

[MariaMaria]

Maria - that would have resulted in very similar set of comments. Perhaps "When to ETTR and when not to" might be a better question.

Agreed

[MariaMaria]

Because we compress the dynamic range that we capture into something we can print or display. That sounds complicated, but in practice, it's what occurs when we use something as simple as a fill light slider in post processing.



That's a common misunderstanding - an 8 bit representation doesn't limit the image to an 8 stop dynamic range any more than limiting the number of steps on a staircase limits the maximum height of a staircase.

I'm going to interrupt here and see if I understand what you all are saying So the DR is the total amount of values the camera is capable of representing but when we capture an image it is limited to the bit range of the sensor?? I apologize if this question seems redundant.

[george013]

This is my first post here. Sorry for my english, it's not my native language. I want to approach the correlation of DR and bits in another way.

Maria,
Split the issue in a hardware and a software site.
The hardware is your sensor. That is able to give an analogue signal when it's hit by light. There is a maximum and a minimum of light that produces a workable analogue output. This maximum and minimum lightquantities determines the dynamic range of your sensor.
Then the software site. That analogue value is converted to a digital value. First will be determined the range of values, this will be the dynamic range of the sensor. And then there will be a value in how many parts it will be divided. That last is a value of "2 to the power of x". I don't know how to write this different. As you can see x can be any value but the range will be the same. Digital cameras usual use 12 or 14 for x and can save that as a RAW-file. This division has absolutely nothing to do with exposure or dynamic range.

The advance of more bits is in editing, you get smoother changes of luminance.

A histogram has a typical division of 256 values, 2 to the power of 8 being the value of a JPG and is nothing more than a graphical representation of your photo.

I hope this helps.

George

[MariaMaria]

This is my first post here. Sorry for my english, it's not my native language. I want to approach the correlation of DR and bits in another way.

Maria,
Split the issue in a hardware and a software site.
The hardware is your sensor. That is able to give an analogue signal when it's hit by light. There is a maximum and a minimum of light that produces a workable analogue output. This maximum and minimum lightquantities determines the dynamic range of your sensor.
Then the software site. That analogue value is converted to a digital value. First will be determined the range of values, this will be the dynamic range of the sensor. And then there will be a value in how many parts it will be divided. That last is a value of "2 to the power of x". I don't know how to write this different. As you can see x can be any value but the range will be the same. Digital cameras usual use 12 or 14 for x and can save that as a RAW-file. This division has absolutely nothing to do with exposure or dynamic range.

The advance of more bits is in editing, you get smoother changes of luminance.

A histogram has a typical division of 256 values, 2 to the power of 8 being the value of a JPG and is nothing more than a graphical representation of your photo.

I hope this helps.

George

Thank you George for explaining it that way. I am able to grasp the difference better now No need to apologize you seem very fluent in English. I am only fluent in one language (English), so I'm sure I would be at a great loss if I tried to communicate in your native language!

[Inkanyezi]

I would be more inclined to call it a correct exposure and the scene fits the dynamic range available. Actually if you look at the histogram with the GIMP there are slight signs of clipping when the view is enlarged. So slight it's hardly worth even mentioning and seems to be mostly in the green channel and some red. These tiny bits could probably be bought in from raw. Once it's in a jpg there isn't really any way of telling if an image is clipped other then the bright end petering away to zero and that needs a large histogram or seeing that it actually falls short of clipping.
...

Wasn't the idea behind ETTR to achieve "correct exposure"?

The problem is a bit more complex than only correct exposure. Often we must consider other aspects as well. When taking hand-held photographs, and when subjects are moving, as well as when the subject is distributed over a range of distances, factors as exposure time and aperture will influence the end result.

So "correct exposure" is the primary goal of exposing to the right, and noise was, and perhaps is, a secondary consideration. Grown up in a tradition of using the Zone System (as outlined by Ansel Adams^*), I see empty space to the left as an invitation to raise ISO. which gives me the opportunity to use a shorter exposure time. Raising ISO shrinks the available dynamic range, effectively moving the histogram to the right, as well as stretching it out over the available space.

So exposing to the right at base ISO is not the same thing as doing it at a higher setting. When ISO is increased in the camera, its amplification of the analogue signal is increased, pushing the histogram to the right. If we use base ISO, we can do a similar thing in post production, digitally imitating what's done analogue with raised ISO. But...

If we take the same picture at base ISO, increasing ISO digitally by crumming the curve upwards, We don't have all those bits in our highlights, just as the ETTR mantra suggests. We will have even fewer bits in the shadows.

The final image will show a minimal amount of clipped highlights when the subject has specular highlights, as this one does. It will most probably have some bits that are pitch black as well. How much of the image to accept pitch black or clipped, is up to judgement.

So of course the RAW file does contain a tad more image information both in highlight and shadow, but none of it will be visible if retained in the final image. It is simply not important for the impact of the image. Here's where "gamma" comes into the equation. Gamma is used to place the most important image information into a tonality that pleases the eye. If we feel forced to get those last bits of data into the jpeg, we can do so, and alter the gamma to slightly higher contrast in the mid-tones to pull the image back into "punch", gaining virtually nothing. Those last bits can also be left alone, clipping, as well as pitch black.

Raising ISO from base is different from doing it digitally in PP, in more than one respect. Analogue amplifying does not cause posterisation. Analogue amplifying pushes the histogram right, causing mid-tones and highlights to contain more data. So one aspect of raising ISO is to use it to push the histogram right, effectively "ETTR", and this process is very much akin to using the original Zone System, as described by Adams. The twentieth century method in the twenty-first century.

Increasing ISO is a measure we can use to push the histogram to the right. It shrinks the dynamic space, but the image can be captured with more data in the important mid-tones with less exposure. How much to increase ISO is a matter of judgement. Here it is raised two stops, effectively decreasing dynamic range by two stops from ten at base ISO to eight stops at ISO 400.

* Blame it on my increasing age or whatever, but when I wrote this, I couldn't remember the first name of Ansel Adams. Having finished the book Dirk Gently's Holistic Detective Agency just yesterday, the only first name that came into my mind was Douglas. I had to rise from my chair and look closer at the print of Sierra Nevada Winter Evening from Owens Valley, California, which hangs on my wall, to get his first name.

[Colin Southern]

Wasn't the idea behind ETTR to achieve "correct exposure"?

Nope.

Have a read of this (from the "guy who invented ETTR")

http://www.luminous-landscape.com/tu...exposure.shtml

Sorry, but my head is starting to hurt from banging it against the brick wall; anyone who over-exposes a black cat sitting on a pile of charcoal by around 4 stops - so they get cleaner data when they reduce it by 4 or so stops back to a black cat (where you can't see the noise) needs their head read in my opinion.

I'm out of this thread.

[Inkanyezi]

So Maria, the essence of my longish post above, combined with the one from George, is that when you raise ISO in the camera, you shrink the dynamic range that your sensor can capture. Base ISO is mostly the lowest figure you have on the ISO scale. For each full step (doubling the ISO) you lose one step of available dynamic range. Often this is said to increase noise, but as previously mentioned, noise often is not very important, although dynamic range more often is.

However, modern sensors have large dynamic range, in some cameras as the newest Pentax and Nikon models up to fourteen stops, while in the time the ETTR was proposed first, cameras only had about eight stops of dynamic range at their base ISO. Modern sensors also have a much lower noise floor, which actually is just another way of expressing dynamic range. What happens when we raise ISO in the camera is that we amplify the signal, and the noise as well. When we raise ISO on a modern sensor, it becomes more like those older sensors that had a higher noise floor, and exposing to the right, in order to better utilise their dynamic range may be more important than at lower ISO settings.

The whole essence of using the histogram to determine exposure is to try holding it within bounds. If the histogram floats into the undefined space outside the left and right boundaries, shadows are blocked and highlights are clipped. In cases where there is much empty space with only baseline to the left or right, ISO can mostly be raised without any loss of important image data and with good rendition of the subject.

Often we will more readily accept blocked lowlight portions of the image than clipped highlights. Therefore in many cases it is important not to let the histogram climb the right wall of the histogram space, but to end just in the lower right corner or close to it. The methods for pushing the histogram left is to expose less (using minus compensation) or decrease ISO. The methods for pushing it to the right is to increase exposure (by plus compensation) or to raise ISO.

But it should be remembered that the histogram or the mantra of ETTR are only tools to use to find out whether we can possibly change one or other of the input parameters.

[dabhand]

For those who may like a detailed (and very technical) discussion on the subject of exposing to the right this weeks Amateur Photographer has an article by Prof Bob Newman - http://www.amateurphotographer.co.uk...e/latest-issue - those outside the UK may be able to get it from a specialist outlet or there is the option of a digital version.

steve

[Inkanyezi]

Wasn't the idea behind ETTR to achieve "correct exposure"?

Nope.

Have a read of this (from the "guy who invented ETTR")

http://www.luminous-landscape.com/tu...exposure.shtml

My misunderstanding doesn't as I see it cause any real loss, but perhaps an even better understanding of how the medium can be used to visualise and perform photography. The maximum dynamic range of the medium is fixed at manufacture, but the actual dynamic range can be changed by the operator to achieve desired goals. When we don't have tunnel vision for noise, but also consider other aspects of the image, accepting a reasonable degree of noise as well as fewer numeric steps in the shadows may be desirable when other properties are more important, as for example exposure time.

I guess I started to misunderstand as a consequence of learning the Zone System first and attempts to apply it to digital photography, where the dynamic range is adapted to that of the subject.

As presented by Michael Reichmann, I perhaps misuse the term, as I instead of exposing more, increase ISO to adapt my sensor's range to the subject, instead of increasing exposure, to get the brightest values in the lower right corner, thus accepting to use the left part of the histogram as well, in order to gain other desirable features, as in the case of the snake a shorter exposure time.

The instances where I would deliberately use a much larger range of the sensor than the scene but increase exposure, in order to get less noise and more tonality in the shadows, to drop those unused bits later in PP, are indeed rare, although I know of, and understand, the possibility and the theoretical advantage. I am fully with Colin in that respect. If I don't need those "richer tones" or less noise, why go to the effort of deliberately exposing more than necessary, having to adjust a too bright image afterwards? It could make sense occasionally, when exposure time can be long and when the promised richness of tones could be important. Most of the time, for me, it would not.

[benm]

That's a common misunderstanding - an 8 bit representation doesn't limit the image to an 8 stop dynamic range any more than limiting the number of steps on a staircase limits the maximum height of a staircase.

I do not understand this statement. Could someone explain this? Thanks.

[dabhand]

I believe what Colin was trying to convey was that there is no relationship between 'bits' and 'f stops' - see http://www.rags-int-inc.com/PhotoTechStuff/TonesnZones/ for a detailed article.

steve

[benm]

I understand that the height of a staircase can be thought of as the dynamic range. And if you have a jpeg image (8 bits) you can have up to 256 tonal values, or steps. But does not the 256 steps translate into 8 f/ stops? How can 256 steps (8 bits) represent more than 8 f/ stops? Below is a table from the CinC tutorial on dynamic range.



I understand that the left column is the bit precision for conversion, not the number of bits in the image, but I don't see how 8 bits can represent more than 8 f/ stops. (At least theoretically. Practically, it is even less).

[Manfred M]

Ben – I think you are confusing two completely different issues; the actual dynamic range that the camera sensor has captured and how that image data is actually generated.

A camera sensor is an analog device that runs through an analog / digital (A/D) converter, so the sensor itself doesn’t actually have a specific mapping to a particular bit value. What the sensor actually does is have is a specific quantity of light, that is represented by a voltage. The A/D converter simply takes a range of voltages and assigns all the readings that fall between these two ranges to a specific value. The higher the bit-depth, the more discrete values are assigned, but it has absolutely nothing to do with what the sensor is capable of measuring. The range (stair-case height) is determined by the sensor design. The step-height is determined by the bit depth used.

As an example; let’s assume (for simplicity sake) that a sensor outputs 0V for pure black and 10V for pure white. For a 1-bit conversion (everything is assigned a value of 0 or value of 1), I could determine any signal between 0V and 5V = 0 (or black) and any voltage over 5V and up to 10 V is assigned a value of 1 or white.

So:

0.00 - 4.99V = (0)
5.00 - 10.00V = (1)

If I decide to upgrade and use a 2-bit A/D conversion, all of a sudden I can handle four distinct values (bit 0 can have a value of 0 or 1 and bit 1 can have a value of 0 or 1), i.e. I can take the same signal and break things down as follows:

0.00-2.49V = (0,0) = 0
2.50 – 4.99V = (1,0) = 1
5.00 – 7.49V = (0,1) = 2
7.50 – 10.00V =(1,1) = 3

So, the dynamic range of the sensor has not changed at all, but with a more sophisticated A/D converter, I can take the same signal and describe 4 distinct states.

I can continue to do this up to whatever bit rate I want, be that 8-bit, 12=bit, 14-bit, 64-bit, etc. The dynamic range of the sensor is still the same, but I keep increasing the number of discrete steps describing the values recorded by the sensor.

Hopefully this clarifies how there is really no link at all between the dynamic range of a sensor and the digital output of the camera.

So if I look at my camera (D800), which at base ISO is rated at 14.4EV, and outputs 14-bit (14-stops according to the chart) data, there is absolutely no conflict here. In theory, to get every bit of performance data out of the sensor, Nikon should have opted for a 15-bit or 16-bit A/D converter.

This also means I can work with a high gamut 14-bit image in post or a 8-bit jpeg; both will represent the lowest voltage range as black and highest as white. A newer sensor might be able to push the limits that can be recorded even highter, and perhaps we will get to work with true 16-bit data, rather than 14-bit data that has been packed with zeros.

[george013]

Manfred,

I understand what you are trying to say. But the above table suggests that there is a correlation between bit precision and dynamic range. There isn't.

George

[Manfred M]

Manfred,

I understand what you are trying to say. But the above table suggests that there is a correlation between bit precision and dynamic range. There isn't.

George

Agreed - we are making the same point. People somehow confuse bit depth with dynamic range.

As you have correctly put it, there is none. The two are completely independent of each other. A value of 0 represents "pure black" and a value of "1" represent "pure white", which is the dynamic range. The number of steps in between the two is irrelevent in defining dynamic range.

Based on what I have read, the human eye can distinguish about 10 million distinct shades, so we have (approximately) 7-bit eyes, but I've also read that we can detect a dynamic range of around 20-stops. Again, there are some major assumptions here including having our irises open up and stop down, plus some "adjustment" lag.

[benm]

If dynamic range can be larger than the bit depth then it would mean that there have to be gaps in the tonal values (posterization)? If this is the case then I can understand how it can be so.

Our eyes may have a dynamic range of 20 stops but not simultaneously. We can only see a portion of that range at any given time (no gaps allowed ).

[Colin Southern]

If dynamic range can be larger than the bit depth then it would mean that there have to be gaps in the tonal values (posterization)? If this is the case then I can understand how it can be so.

Our eyes may have a dynamic range of 20 stops but not simultaneously. We can only see a portion of that range at any given time (no gaps allowed ).

There are ALWAYS gaps in the tonal range because it's a digital process (just like there are no 1/2 steps on a staircase) - it's just a case of "are those gaps big enough to cause a problem?". Posterization occurs when the answer to that question is "yes".

In camera terms, that's not particularly related to the dynamic range - if you had a large dynamic range and a very small number of steps then it could be a problem - but as things stand, it's only a problem if you're trying to use some of that extreme low-end data, and with a modern camera that's only likely if the original image was grossly under-exposed or the dynamic range of the scene that you needed detail from was extreme.

In the real world, most people will be dealing with 14 bit cameras and looking no deeper for detail than about 9 stops down (detail from dark detail that's in the sun's shadow from an exposure shot with a 2 stop safety margin), and even that would have 16 levels available if my math is correct (which is doubtful this time of the morning!)

[george013]

If dynamic range can be larger than the bit depth then it would mean that there have to be gaps in the tonal values (posterization)? If this is the case then I can understand how it can be so.

Our eyes may have a dynamic range of 20 stops but not simultaneously. We can only see a portion of that range at any given time (no gaps allowed ).

Ben,
There is no relationship between dynamic range and bit-depth. You won't come further if you don't see that. Forget the table you refered to, it's wrong.
A stop is an optical unit, bit-depth an arithmetical unit.That they are both sort of binary based, doesn't make them equal.

I don't know about the 20 stops of our eyes but I know there is a lot of PP in our head.

George

[Inkanyezi]

The over-simplification that connects the concept of bit-depth with dynamic range is a bit confusing. In short, with another ladder than the simple n² steps for f-stops, the concept of exposing to the right can still make sense. It does not matter exactly how many levels we have in the highest region or the lowest, but there is always more data in a higher, brighter, zone than in any lower, darker, zone. This is the fundamental idea behind ETTR.

You get less noise when pushing your exposure to the right by exposing more, but it does not make sense to always do so, as other factors than noise, particularly almost invisible noise, may be more important. ETTR is one tool that can be used, sometimes. The issues in photography are very often other than getting maximum tonality and minimal noise in the darkest parts of an image.

However, in conditions where noise might be a problem, as when your dynamic range shrinks with high ISO, it does make sense to use all available dynamic toward the brighter end of the histogram.