Dynamic Range - What's the trade off?

[pnodrog]

I don't have a Fuji and am not interested in one. I'm just surprised how arguments are set away.

The example of Robin might not be the best. There is hardly clipping in it. Highlights doesn't mean clipping by definition.
When there's clipping in the image, it's impossible that to correct in pp and get back details that where not there. The only way to get back details in the clipping area is exposure correction.
As I stated many time before, one can not see a RAW-file as an image. You see the result of the converter.

George

George as far as I understand it the adjustment is done by reducing the exposure and then applying a correction curve to it after exposure. More or less what I have done for years when I was worried about loss of highlight detail. ( shoot RAW -0.3 or -0.7 exposure compensation followed by curves adjustment in PP) I do it very seldom now due to the good Dynamic Range of the D800 and better skills honed due to practice restoring ETTR exposures.

It maybe a useful option but nothing special. This thread just proves you need to know your camera to get the best out of it.

[Black Pearl]

Thats a fantastic and detailed post Dave thanks for taking the time.

I was thinking about this last night and started looking at it this way. If we work on the basis that the actual sensitivity the sensor is the equivalent to 200iso then as you set a higher iso on the camera really all you are doing is letting less light than is needed to produce a correct exposure hit the sensor meaning the camera then has to apply gain to 'correct' it. On that basis when using 800iso then all the DR modes are doing is altering the way in which the gain is applied - regardless of the mode the camera needs to apply two stops (400%) gain to get the exposure looking correct. At DR100 the camera is applying 400% gain uniformly across the entire image whereas at DR400 the camera is manipulating that gain using smaller amounts on the highlight regions and larger amounts in the darker regions and balancing them with mid tone gain.

Looking at it this way the camera isn't under exposing the image when you use the DR modes it simply taking the already under exposed information and manipulating it differently.



George you've lost me. The amount too clipping (there or not) in my examples is not relevant in any way.

[george013]

George as far as I understand it the adjustment is done by reducing the exposure and then applying a correction curve to it after exposure. More or less what I have done for years when I was worried about loss of highlight detail. ( shoot RAW -0.3 or -0.7 exposure compensation followed by curves adjustment in PP) I do it very seldom now due to the good Dynamic Range of the D800 and better skills honed due to practice restoring ETTR exposures.

It maybe a useful option but nothing special. This thread just proves you need to know your camera to get the best out of it.

Exactly. And exposure, light or photons gathered by the sensor, can only be changed with A or S. One can play in PP with an exposure correction, but that's not possible for the clipping parts.

George

[george013]

George you've lost me. The amount too clipping (there or not) in my examples is not relevant in any way.

I'm afraid you didn't understand the meaning of "dynamic range extension". It's only ment for those situations where the dynamic range of the scene is bigger as the dynamic range of the sensor. Resulting in clipping. If you don't have clipping, then the scene is within the dynamic range of the sensor. And nothing happens.

George

[Black Pearl]

I'm afraid you still don't understanding the original question or the general discussion in this thread or that the files I took don't need to include actual clipping to demonstrate the way the camera is manipulating the data. The thread is not about specific files its about how the camera implements the DR Expansion modes and what impact it has on the quality. I feel both Dave and myself have provided sufficient information to answer the OP question and I personally have a better understanding of how my camera operates.

In a nutshell for you - if you take a photograph with a Fuji using the DR400% mode and even if the underlying data doesn't have actual clipping the camera still tones down the highlights to present a more uniform image. It does so by altering the way the gain is applied to the data and while that does have an impact on the noise levels it is very well controlled and still real-world usable.

[Dave Humphries]

Hi George,

I'm afraid you didn't understand the meaning of "dynamic range extension". It's only meant for those situations where the dynamic range of the scene is bigger as the dynamic range of the sensor. Resulting in clipping. If you don't have clipping, then the scene is within the dynamic range of the sensor. And nothing happens.

This might be true for the way 'Auto' ADL works on Nikon cameras, but I don't think it is true of the Fuji implementation.

And exposure, light or photons gathered by the sensor, can only be changed with A or S.

I do kinda get where you're coming from with this, but what I think you're overlooking is that regardless of what is in the pixel's sensels (in photon count), it is the how that data is extracted and stored (for later PP) that effectively determines the ISO - and ISO is part of the exposure triangle. Dave's RAWDigger graphs show this to be the case with the differences between the three '800 ISO' DR100, DR200 and DR400 plots.

It is also interesting to see that there is some clipping in the DR100 shots (R, G & B), less in the DR200 shot (G only) and none in the DR400 shot - I think Robin's shots do provide us with good enough examples to prove the discussion's concepts.

However, even if there were no clipping, the Fuji system relies on the photographer not to use it when it is not required; e.g. for a scene of limited contrast range that is within sensor's native DR.
The areas of Robin's images which are not into the clippers were still under exposed by one or two stops, but then brought up in ACR to be seen as correctly exposed.


All,

What I take away from all this is that with ISO-less cameras, as long as the highlights are not under exposed *, it shouldn't matter (if you're skilled at PP) whether you recover in PP or in camera by a higher exposure**.

* What this should not be interpreted as is carte-blanche to under expose in low light and recover in PP, since that will cause an increase the noise

** In this instance, you may lose highlight detail, although doing so may be of no consequence; e.g. small specular reflections, but the noise in the shadows should be similar as Dave has shown from Robin's shots.


Personally, I have always tried to ETTR for best noise performance, but never at the expense of losing highlight detail that matters to me. Then, in PP, I have manipulated the Levels or Curves to achieve what looks like a well exposed shot, but one that usually contains good data in say, the skies. I haven't used Nikon's ADL on my cameras, since I found it was too easy to leave it on when I moved to a different scene - in practice I was/am doing it manually with EC anyway, by careful attention to the RGB histogram and blinkies (noting their 8-bit limitations) and re-shooting when necessary to preserve highlight detail. It may well be that I am losing out when that's not possible, for example, shooting planes at an air show - in future, it seems I should be using ADL.


My thanks to all the main players in this thread, Dave

[John 2]

I'm afraid you still don't understanding the original question or the general discussion in this thread or that the files I took don't need to include actual clipping to demonstrate the way the camera is manipulating the data. The thread is not about specific files its about how the camera implements the DR Expansion modes and what impact it has on the quality. I feel both Dave and myself have provided sufficient information to answer the OP question and I personally have a better understanding of how my camera operates.

In a nutshell for you - if you take a photograph with a Fuji using the DR400% mode and even if the underlying data doesn't have actual clipping the camera still tones down the highlights to present a more uniform image. It does so by altering the way the gain is applied to the data and while that does have an impact on the noise levels it is very well controlled and still real-world usable.

I understood that from your reply in chapter 1 Robin but I'm not one to get I the way of an in depth discussion.

[dje]

Thats a fantastic and detailed post Dave thanks for taking the time.

I was thinking about this last night and started looking at it this way. If we work on the basis that the actual sensitivity the sensor is the equivalent to 200iso then as you set a higher iso on the camera really all you are doing is letting less light than is needed to produce a correct exposure hit the sensor meaning the camera then has to apply gain to 'correct' it. On that basis when using 800iso then all the DR modes are doing is altering the way in which the gain is applied - regardless of the mode the camera needs to apply two stops (400%) gain to get the exposure looking correct. At DR100 the camera is applying 400% gain uniformly across the entire image whereas at DR400 the camera is manipulating that gain using smaller amounts on the highlight regions and larger amounts in the darker regions and balancing them with mid tone gain.

Looking at it this way the camera isn't under exposing the image when you use the DR modes it simply taking the already under exposed information and manipulating it differently.

No problem Robin, it's been an interesting excercise. I think what you have said here is a pretty reasonable way of summarising the situation.

Just out of interest, I had a quick look at Active D Lighting on my Nikon, in raw mode. It seems to behave in a similar way (ie under-exposing) but not by changing ISO. It simply makes the shutter speed faster in Aperture mode and the Aperture smaller in Shutter priority. In Manual mode, you still have control over both shutter speed and aperture but the metered exposure indication is shifted. So if you adjust ss and aperture to get what looks like a good exposure, you'll actually get some underexposure.

Dave

[dje]

I'm afraid you didn't understand the meaning of "dynamic range extension". It's only ment for those situations where the dynamic range of the scene is bigger as the dynamic range of the sensor. Resulting in clipping. If you don't have clipping, then the scene is within the dynamic range of the sensor. And nothing happens.

George

George no matter what the dynamic range of the scene is, you can get clipping if you don't get the exposure right. If the DR of the scene is greater than that of the camera, you can still set the exposure to avoid clipping but it means that the shadows dis-appear into the noise or are totally black. ie all detail is lost in the deepest shadows. DR extension is about bringing up those shadows in a suitable way, without damaging the highlights.

Dave

[Inkanyezi]

I think the whole issue of exposure and processing, as well as dynamic range optimisation is very difficult to discuss as long as we don't share a mutual language to convey ideas about all properties. We also must be specific about what part of the entire process we discuss at any particular instance. As it is now, we talk about ISO, applying double standards (I didn't say hypocrisy), as ISO in digital photography is an evasive property. We had a similar problem also in photography on film, although Ansel Adams resolved that long ago. When we applied "push processing", we simply did not adhere to the ASA/DIN standards, and the same happens when we raise the ISO figure in the camera. We cannot change the sensitivity of the chip, but we can manipulate the output. This can be done either by altering the analogue output of the sensor, or by digitally manipulating the output after A/D conversion. The camera manufacturer is free to do either; change virtual sensitivity (marked ISO in menus and controls) modifying digital or analog gain (or both). Rumor has it that some manufacturer actually does both.

When the ISO figure is increased by raising analog signal level, amplifying the sensel signal, noise is also raised alongside image impression, and the brightness level moves up the voltage range, which causes digital information in low light levels to get more steps on its ladder, smoother digital resolution. The analog signal can only be increased before A/D conversion, before the image is registered. It is a boon to understand that analog amplification cannot change the sensitivity of the chip, only amplify its output.

When the ISO figure is increased by raising digital signal level, amplifying digital output, noise also comes along, and the brightness level moves upwards, but there is no increase in signal resolution, the ladder will have fewer steps, coarser. Amplifying digital output can be done anywhere in the processing chain between registering data from A/D conversion to presenting the image to an observer. The observer will not know whether it was done right after the A/D converter in the camera, by a computer program, in the graphics board in the computer, or on the computer screen. There simply is no way to tell, visually. In digital processing, the steps of the ladder can be increased, synthetically, by adding data that was not there from start, a process called sample rate conversion. This is similar to what is done when upscaling a digital image by upsampling. Purists may object that it is not photography, as the added data was not acquired by light impression of photons.

DRO, optimising dynamic range, is not a simple concept, but there are hopes of applying DRO at the flick a switch. It can be useful to know how DRO is applied; what idea of optimising and method of implementation that applies. Different concepts of how to optimise should perhaps be resolved? Maybe the automatic system of the camera will not comply with your own idea of what to optimise?

An image chip has a maximum dynamic range at base ISO. It is not possible to get maximum dynamic range at other ISO settings than base ISO. However, optimisation does not imply that maximum dynamic range should always be used. If the scene has a dynamic range of less than the maximal range of the chip, we can decrease the dynamic range to register, below the maximum of the chip. One concept of "ETTR" does this by moving captured data to the right of the histogram, in order to avoid noise in the darker levels of the image and get more useful data. This is one idea of optimisation. We had lots of flak fired in the past on that subject.

Optimising dynamic range may imply that available scene dynamic range is spread over the range to register by the camera. This idea implies adapting ISO in the camera by amplifying the analog signal until it barely covers the range of the scene, provided the scene has less range than the base ISO of the chip. This corresponds to the ideas of Ansel Adams, when he invented the zone system. In fact, his first book The Negative was about optimising dynamic range, by exposing and processing to cover the scene dynamic range. It is good read also in the digital age.

In many discussions of optimisation of dynamic range, the aspect of adapting it to a scene with less range than the chip is completely forgotten, considering only largest possible dynamic range.

One aspect that I did not take up above is noise level. The ETTR idea was about noise level, considering noise to be a problem. I don't regard noise as a problem, but as a property, to be used just as other properties of the medium. Noise is generated mostly by things outside our control, electric noise from various sources in the signal chain, starting at the sensel itself. Anyone who bothers about noise might prefer using a lower ISO to subdue it, even if a high ISO would cover the dynamic range of the scene.

Here's where DRO schemes may collide with our imaging desires. A DRO idea that implies raising the darker levels of an image to more visible, brighter levels, also will raise noise in these levels of the image, compared with noise levels acquired by shooting without DRO, at scene dynamic ranges that approach the range of the chip. A typical situation may be an indoor scene with a window where brightness levels are far higher. Our choices are multiple. We can accept that anything outside the window may clip, concentrating on the tonality indoors, to get good rendition of objects inside the room. We can also try to get as much as possible of the outdoors scene well rendered, at the same time as getting indoors brightness levels sufficient to render objects well, at the price of more visible noise in the indoors regions. This is what Active D-Lighting does, and it corresponds to most people's concept of optimising dynamic range. It implies that the entire frame is shot with less exposure than in the former case, with a tone curve that is different from "normal" processing.

The trade-off of a DRO method that applies less exposure for high contrast scenes with altered tone curve, is that at bright levels contrast will decrease, and at dark levels noise will become more visible.

There is yet another possibility to visually present a larger dynamic range to the observer, which implies tone mapping of the image. This is done by some software, but so far, I don't know of any camera that would do it in its jpeg processing. Some modern sensors may capture up to 14 stops above the noise threshold, thus enabling HDR imaging with one single exposure. Just as the previously mentioned DRO method (Active D-Lighting), some areas will display more noise, but the image may look nicer, as its local contrast is raised by the tone mapping process.

[xpatUSA]

As I stated many time before, one can not see a RAW-file as an image. You see the result of the converter.

Off topic but, as I have also said many times before, that statement is completely false. Or, it is true only in the restricted sense that one can also "not see" a TIFF or a BMP file, because one can "not see" actual columns of data as an image.

What you see below is what the sensor got:



This is an image made with absolutely no conversion - that is to say there is no "the converter" used to make this image. The raw file values are rendered directly to the screen driver with only scaling and gamma correction but with no conversion to RGB.

I offer this post as conclusive proof that "a RAW-file" can be seen as an image, just as a TIFF file can be seen as an image. I expect that there could be some argument from someone, so may I request that any reader who agrees - please mark this post as helpful.

[george013]

Off topic but, as I have said many times before, that statement is completely false. Or, it is true only in the restricted sense that one can also "not see" a JPEG-file, because one can "not see" actual columns of data as an image.

What you see below is what the sensor got:



This is an image made with absolutely no conversion - that is to say there is no "the converter" used to make this image. The raw file values are rendered directly to the screen driver with only scaling and gamma correction with no conversion to RGB involved.

And 'll say again and explain again.
An image as we see and work with is a RGB raster image. A RAW file is a R,G,B image. Again, pay attention to the comma's.
What you show us is a RGB raster image where each sensel containing only one colour is transformed to a 3-byte RGB pixel. The 2 missing colors are given the same value as the existing colors, so showing a gray.
A JPEG is a diskfile containing a compressed image of that image. Loading it in memory only means to decompress it

So, we both said it again.



George

[Inkanyezi]

What you show us is a RGB raster image where each sensel containing only one colour is transformed to a 3-byte RGB pixel.

I doubt that is the case, as it is a Foveon sensor where any pixel has all three values. However, just as in all other electronic imaging, sensel responses are linear, so conversion is needed to create an image with correct tonal values.

[xpatUSA]

And 'll say again and explain again.
An image as we see and work with is a RGB raster image. A RAW file is a R,G,B image. Again, pay attention to the comma's.
What you show us is a RGB raster image where each sensel containing only one colour is transformed to a 3-byte RGB pixel. The 2 missing colors are given the same value as the existing colors, so showing a gray.

Hi, George. It seems again that you are describing a sensor with a Bayer CFA. However your original statement does not make that clear and there are other types of raw file that are not in the form of RGBG - with or without commas.

So it is that my Foveon raw files of type X3F are in the form of TMB (top layer, middle layer, bottom layer). Nevertheless they are still RAW files and can not be called anything else. So, no "transformation to a 3-byte RGB pixel" took place in the rendition on-screen of my posted raw image.

A JPEG is a diskfile containing a compressed image of that image. Loading it in memory only means to decompress it

I edited "JPEG" out of my post earlier, so as to avoid this kind of red herring (obfuscation). Let's stick with un-compressed TIFF or PNG.

So, we both said it again.

George

Yep. By the way, I may be beginning to get your point, I think. Are you saying, in other words, that normal Bayer-CFA raw image data is in four columns but a raster pixel is in three columns? So, with such a raw file, something has to be done to turn four into three? Not applicable to my raw files, of course.

[george013]

Well, we where discussing the Fuji. I was waiting for this comment. Urban was first.
TIFF is holding a RGB rasterfile. The value of the pixels are fixed. Except when compressing them.

George

[Black Pearl]

I've got my popcorn and a beer - hopefully this should be as entertaining as ever......

[Dave Humphries]

I've got my popcorn and a beer - hopefully this should be as entertaining as ever......

Think I've seen this one before

[xpatUSA]

Well, we where discussing the Fuji. I was waiting for this comment. Urban was first.

If you only meant the X-trans, I must most humbly apologize. I thought that "one can not see a RAW-file as an image" was a general statement.

https://upload.wikimedia.org/wikiped...lourfilter.svg

TIFF is holding a RGB rasterfile. The value of the pixels are fixed. Except when compressing them.

And except when they are holding one layer only.

Hmmm . . can one layer of RAW data be seen as a image? . .

The problem with this discussion is because a statement was made that was apparently general - but was not so in fact.

Misunderstandings then followed.

It's all your fault, George