Dynamic Range War

[CanineCandidsByL]

Monitors have more DR than prints. Some may even exceed sensors.

Sorry, but thats not true. Some extreme monitors might but generally if you look at a monitors specs it will show you differernt.
For example, I looked up one monitor and it shows a 1:20,000 (and you can find ones up to a 1:1,000,000), but in the specs it says 1:1000 typical.

The backlight intensity can vary which gives you the difference between the two specs. So for a movie that moves between bright and dark, you might get 1:20,000 difference between the darkest pixel in a dark scene and the brightest pixel in a bright scene. However for a stationary picture, your limited to the amount of filtering the foward layer can provide...in this case 1:1000.

For those who haven't left already, how many stops are in these ratios? Remember a stop is a doubling of light and that 1 stop of DR represents 2x between the darkest and lighest. So we start with DR of 1 equal to 1:2 and we double each time....

Stops Ratio
1 1:2
2 1:4
3 1:8
4 1:16
5 1:32
6 1:64
7 1:128
8 1:256
9 1:512
10 1:1024 (typical for most monitors)
11 1:2048
12 1:4096
13 1:8192
14 1:16384
15 1:32767
16 1:65536
17 1:131072
18 1:262144
19 1:524288
20 1:1048576 (the largest "nonsense" number I have ever seen for a monitor)

I should make a special mention for LED monitors. Instead of one light source for the whole monitor, they have LEDs that can be individually adjusted brighter/darker. As such, you actually might get those 1:1000000 ratios on the screen, but a single LED controls a group of pixels so a white pixel next to a black pixel would generally be back at that 1:1000 figure.

TMI? ;D

 

[keithfullermusic]

This image shows why we would like more dynamic range in a camera. I took this on a bright day, and the camera exposed for the bright sky. I then switched to manual exposure and exposed for the person, but kept the image as a example.

A higher dynamic range might let you get good detail from both the bright and the shadow areas, without enough dynamic range, you must chose the area you want to be able to show the detail.

Black and white film typically had very good dynamic range, and you routinely got good prints from a bit of under or over exposure. Digital is not so forgiving, its because there is less dynamic range.

Most of the discussion involves how much range you can theoritically extract from a raw image, and is not necessarily related to the quality of the final image.

If DR was that high then they wouldn't make any money selling flashes!!!

Also, You can just use grad filters for things like this. Granted, it wouldn't be ideal right here where the top half of the guy would be darker than the lower, but grad filters solve this problem with ease when taking landscape shots. Polarizers also do a good job of darkening the sky and water more than other things, so that might help in situations like this.

I bet you could lighten the shadows in LR/Aperture and get tons of info back on the darker parts also.

Granted, more DR would be sweet, but there are some ways around it.

 

[helpful]

For the record, slide film has a smaller dynamic range than negative film.

However, it still held at least as much "data" as did negative film.

Therefore, there was much more detail in the image coming from a slide, but exposure latitude was not as forgiving. Blown highlights, and lost shadows were more likely with slide film. Overexposure in particular is much less of a problem with negative film, because there was a lot of headroom with negative film. Not so with slide film.

The way this compares to digital images is as follows:

Approximately the brightest 1/2 of the image data (i.e., the right side of the histogram) represents one stop. Therefore, the most image detail is in this area. Subtle variations between brightness are easily discernable.

The darkest 1/2 of the image data is divided up as if it was another image, like this:

* the brightest 1/2 of the remaining 1/2 of the image data (i.e., the upper half of the bottom half of the histogram) contains another stop of brightness data. There is slightly less detail, since not as much data is used to represent one stop of light.

The remaining quarter of the image data is divided up again, recursively:

* the brightest 1/2 of the remaining 1/4 of the image data contains another stop of brightness. There is significantly less detail, etc.

Let's assume that your image data is stored as 16 bits per pixel, and the primary part of your image is exposed at EV 16.

Then the top 8 bits contain a detailed view of the brightest one stop of the image from EV 16.0 to 16.9.
The next 4 bits contain a moderately detailed view of the next brightest stop of the image from EV 15.0 to 15.9.
The next 2 bits contain a poorly detailed view of the next brightest stop of the image from EV 14.0 to 14.9.
The final 1 bit contains a very low detail view of the darkest stop of the image from EV 13.0 to 13.9.


This is how it works, simplified, with "linear gamma."

All modern cameras use non-linear gamma systems to expand the dynamic range of linear gamma from the maximum of four stops to many more stops of dynamic range.

But the principle is the same.

If you want to get the most detail from your images, then the major area of your image's histogram should be towards the right.

Try taking a photo of a subject with little dynamic range, like a a square foot in the middle of a field of green grass. If the image is underexposed, the histogram will make a narrow band towards the left side. The size of the band represents the detail recorded from the grass.
If the image is properly exposed, the band will be a little bit larger showing that more detail is being recorded in the upper area of the histogram which is more detailed because more bits are used to record each stop of light.

 

[Marsu42]

If you want to get the most detail from your images, then the major area of your image's histogram should be towards the right.

Thanks for pointing that out, I only discovered this fact after some try and error - maybe they should have put a sentence like this in the manual

 

[neuroanatomist]

If you want to get the most detail from your images, then the major area of your image's histogram should be towards the right.

Thanks for pointing that out, I only discovered this fact after some try and error - maybe they should have put a sentence like this in the manual

The concept is termed ETTR.

 

[Marsu42]

The concept is termed ETTR.

While it might not make me seem like a pro (hey, I am not!): I didn't know about that, thanks again dr. neuro!

The only thing left for me to wonder when shooting at low light is if it's better to have a properly exposed histogram at higher iso or a histogram that leans to the left at lower iso. Using lr4 and its smart shadow recovery, I'm tending towards the latter with my aps-c sensor, because higher iso than 800 really ruins the picture while 1ev underexposure does not.

 

[helpful]

The concept is termed ETTR.

While it might not make me seem like a pro (hey, I am not!): I didn't know about that, thanks again dr. neuro!

The only thing left for me to wonder when shooting at low light is if it's better to have a properly exposed histogram at higher iso or a histogram that leans to the left at lower iso. Using lr4 and its smart shadow recovery, I'm tending towards the latter with my aps-c sensor, because higher iso than 800 really ruins the picture while 1ev underexposure does not.

That is a question that there may never be a definitive answer for.

In my experience it depends on the camera. Some cameras whose high ISO modes are just "software enhanced" are better used at lower ISOs with slight underexposure, and then pushed on the comupter with software that is better designed and can increase the brightness without resulting in as much nosie as the in-camera "software-enhanced" ISO boosting.

For other cameras that really do have the ability to increase their light sensitivity, it is absolutely better to increase the ISO to get more bits of detail in the file, and then try to decrease noise later. Underexposing reduces the actual amount of data that is captured in the scene. So in an ideal world, you would get the proper exposure by increasing the ISO, and then reduce the nosie in post-processing, and your image would have higher quality than shooting underexposed and then boosting the ISO with software.

It all depends on whether the ISO level that you are shooting at is provided by the intrinsic analog/physics capabilities of the image sensor physics (in which case you should shoot at high ISO and reduce noise afterwards), or by the camera's image processing (in which case you should shoot at lower ISO and increase brightness afterwards).

Boosting ISO with software always reduces the amount of usable data in the image (which is why the Nikon D4 only has 6-8 bits of DR at ISO 200,000, because it is software boosted). So that is what you always want to avoid.

 

[TrumpetPower!]

I'd caution strongly against expose to the right and even bothering with thinking about linear encoding and the like. Even if that's still what's going on in the silicon, it's been ages since that sort of folk wisdom has had any practical application. Unless you've got a very specific, uncommon, awkward, and carefully-crafted workflow, you're just going to risk blowing your highlights and wind up with unnatural and weird-looking tonal and color shifts.

Expose properly, ideally with a well-calibrated incident meter.

If you've still got crushed shadows and blown highlights in critical areas of the image, you either need better light or you need to go to HDR -- and that's assuming that the crushed shadows and blown highlights are a problem in the first place...the kinds of photography where it's a problem but you can't either fix the light or use HDR are basically nonexistent.

Don't forget that there's a great deal more DR to be had in any well-exposed RAW image than what comes right out of the converter with the default settings. Much more, in fact, than any of the various numerical tests would lead you to believe. And, unless your printer is too big to fit on a tabletop, noise simply isn't a factor any more. Those whose printers take ink by the gallon have to worry about it, but they also generally know how to capture good exposures such that noise again doesn't become a problem or is at least manageable / acceptable.

And that's why the whole brouhaha over the wider dynamic range of the D800 over the 5DIII is meaningless. In the real world, you're never going to find yourself in a situation where the 5DIII has insufficient DR but the D800 is good enough. That minuscule set of scenes where it could theoretically apply still requires either better light or HDR for proper results, even if you've got the D800 in hand.

It's also why Canon went the better route this time 'round: the 5DIII's non-sensor qualities (AF, FPS, etc.) are significantly better than the D800's, and those qualities have the potential for significantly more improvements in image quality than just a few extra megapickles. At least, they do if you're shooting something other than dollar bills taped to brick walls....

Cheers,

b&

 

[helpful]

If you want to get the most detail from your images, then the major area of your image's histogram should be towards the right.

Thanks for pointing that out, I only discovered this fact after some try and error - maybe they should have put a sentence like this in the manual

The concept is termed ETTR.

Great point, I would like to add a link a really cool article on the subject as well:

http://www.luminous-landscape.com/tutorials/expose-right.shtml

In summary,

"For Maximum S/N Ratio [i.e., image quality]

"The simple lesson to be learned from this is to bias your exposures so that the histogram is snugged up to the right, but not to the point that the highlights are blown. This can usually be seen by the flashing alert on most camera review screens. Just back off so that the flashing stops."

There are photos of the histograms in the article.

Note that for a dark subject, you still need the image to be darker than for a bright subject, so it is overly simplistic to say "always have the histogram snugged up to the right." But to get the most details out of all the data that is available in the light coming from the subject, that is the way.

 

[Aglet]

This image shows why we would like more dynamic range in a camera. I took this on a bright day, and the camera exposed for the bright sky. I then switched to manual exposure and exposed for the person, but kept the image as a example.

A higher dynamic range might let you get good detail from both the bright and the shadow areas, without enough dynamic range, you must chose the area you want to be able to show the detail.

Black and white film typically had very good dynamic range, and you routinely got good prints from a bit of under or over exposure. Digital is not so forgiving, its because there is less dynamic range.

Most of the discussion involves how much range you can theoritically extract from a raw image, and is not necessarily related to the quality of the final image.

This image is a good example of how to make use of the full DR of a sensor - altho it hasn't yet been processed to do so for final ouput if one wanted to achieve that look.

You expose for the hilites so they're not blown. Then you tone-curve adjust to lighten the dark areas. There are numerous ways to do this but Adobe's FILL LIGHT is one of the simplest and most effective.

Where the DR limitations come in, shadows at low iso, is NOISE.
When you start to brighten those shadows you may also start to see more chroma noise show up in those areas. If the noise is random it's more acceptable and easier to minimize its appearance.

If the noise has a pattern to it, banding, cross-hatching or similar, then it's very difficult to impossible to remove the appearance of this noise.

This is where the Canon vs Sony-Nikon sensor argument arrises.
One company's sensors have more pattern noise at low iso shadows than the other company's, thus limiting the ability to boost shadows and achieve an effective HDR image from one exposure.

Other than that, they all make pretty good cameras and each has their respective compromises. you choose the one that works the way you need it to.

 

[helpful]

I'd caution strongly against expose to the right

Expose properly, ideally with a well-calibrated incident meter.

If you've still got crushed shadows and blown highlights in critical areas of the image, you either need better light or you need to go to HDR -- and that's assuming that the crushed shadows and blown highlights are a problem in the first place...the kinds of photography where it's a problem but you can't either fix the light or use HDR are basically nonexistent.

Don't forget that there's a great deal more DR to be had in any well-exposed RAW image than what comes right out of the converter with the default settings.

Good points, and actually you have explained very well what the point of ETTR is, which is to maximize the potential of camera sensors, which are capable of far more than what is offered by default exposure settings. The camera sensors are so good that the dynamic range of most scenes is far within the maximum DR capabilities of the sensor. So photographers are faced with a question--use zero exposure compensation, or try to manually expose to get the most detail? Due to the way that data is recorded (2,048 bits of data for the right side in the brightest stop of light), there is much more data recorded on the right side of the histogram.

http://schewephoto.com/ETTR/index.html

There is an amazing example on that page which shows that the little tiny blip on the right side of the sensor (way, way, overexposed, and extreme ETTRing) actually has as much image detail as almost the entire image histogram. That's not what anyone should do. The point is just to show how much more data and details are being recorded for any part of the image that is on the right side of the histogram, compared to the left.

 

[TrumpetPower!]

http://schewephoto.com/ETTR/index.html

There is an amazing example on that page which shows that the little tiny blip on the right side of the sensor (way, way, overexposed, and extreme ETTRing) actually has as much image detail as almost the entire image histogram. That's not what anyone should do. The point is just to show how much more data and details are being recorded for any part of the image that is on the right side of the histogram, compared to the left.

Actually, that page is an excellent example of why not to do ETTR.

Look at the three adjusted histograms. Notice that tall spike on the right on the rightmost histogram? See how it's the only part of the histogram significantly different from the other two? And how it goes all the way to the top?

Even though that spike isn't all the way at the right edge, it still represents saturated, blown-out pixels. All that's happened is that ACR has uniformly reduced the blown pixels to a still-uniform value less than maximum.

Sure, the shadows are cleaner. But I bet a 100% crop of that ``Prairie'' sign would show much more detail in the properly-exposed version than the overexposed one. Were that a wedding dress, Mr. Schewe's smartypants exposure hijinks would just have made the mother of the bride very angry indeed. Even though his histogram showed a "good ETTR" exposure.

And have a look at the waterfall, too. Sure, he was able to recover a good amount of tonality, but the colors are posterized to a ridiculous extent. That exact same sort of posterization is going on in all the other overexposed highlights, with the degree of posterization proportional to the amount of overexposure.

In other words, using ETTR means all your specular highlights will be either devoid of color or have that same sort of severe posterization. Now, granted, the definition of specular highlights is that they get blown...but they'll be much bigger in area and the transition from colorful to blown will be much more abrupt and less colorful. You're basically taking a sledgehammer to your specular highlights, when they really should (in my opinion) remain light and delicate.

By all means, if you like the ETTR look, especially if you're shooting sleeping black cats at the bottom of a coal mine, go for it. But, when that cat wakes up and you want to capture the glint of the candlelight in her eye...use ETTR if you want the glint to be a hard-edged white outline, and expose properly if you want it to look like a candle flame.

Cheers,

b&

 

[sarangiman]

* Current DSLRs are in the 10-12 stop range. (Note: the newest FF bodies from Canon and Nikon haven't been tested yet.)

* Print film has 9-14 stops depending on emulsion.

dtaylor: Print film is more forgiving, yes, but I've always found it hard to compare the stops of DR in print film vs. digital b/c it boils down to: how much noise are you willing to accept in the shadows? If you look at Roger Clark's treatise on DR (http://www.clarkvision.com/articles/dynamicrange2/), negative film falls apart so quickly that its acceptable DR is significantly lower than that of even earlier generation DSLRs & almost on par w/ some slide films (though with Velvia, e.g., the actual signal is very low so that, even with lower noise, it may be hard to extract that signal without a drum scanner). However, latitude with slide film is terrible; I'm always afraid of clipping shadows or blowing highlights. With negative film, I overexpose a sunset by 3 stops & still retain color around the sun. But the shadows are still starving for exposure & are just obliterated by noise! For example, here's a shot overexposed by 2 2/3 stops (as compared to what Evaluative Metering thought the exposure should be on a EOS-3):

Ektar did well, considering the DR of the scene (the sun is still high up in the sky). But those rocks are incredibly noisy upon closer inspection.

So it's my opinion that your estimation of the DR of negative film is highly dependent upon your subjective opinion of acceptable SNR in the shadows.

Which is why DXO attempts to standardize measurements by setting that acceptable SNR to 1 in their measurements. But, like you, I have my doubts of DXO measurements when they claim the D800 has 1.4 stops more DR than the D4, normalized or not. At a pixel level (not normalized), DXO claims 13.24 stops DR for the D800, but only 13.1 stops for D4 normalized (so, less at the pixel level). That just doesn't make any sense.

True, but print film quickly fell apart on the shadow side.

Exactly. You said it yourself

"For Maximum S/N Ratio [i.e., image quality]

"The simple lesson to be learned from this is to bias your exposures so that the histogram is snugged up to the right, but not to the point that the highlights are blown.

Agreed, but this brings up another issue I've had with most RAW converters/image processing software I've had for a while -- when you ETTR, even without blowing out channels, you quickly desaturate bright regions like skies. Software doesn't make it easy for you to recover those tones, which is why I've often found myself being careful about how much I ETTR when I want saturated skies in my final photo. Lightroom 4 is changing that with their 'highlights' & 'whites' slider, which now allow you to really pull back color/tones from bright regions of your photograph. Aperture/Photoshop has allowed you to do this, with limited ability, with their 'Highlights/Shadows' tools for some time now... but I never found it to be enough or as good as LR 4 now.

Where the DR limitations come in, shadows at low iso, is NOISE.
When you start to brighten those shadows you may also start to see more chroma noise show up in those areas. If the noise is random it's more acceptable and easier to minimize its appearance.

If the noise has a pattern to it, banding, cross-hatching or similar, then it's very difficult to impossible to remove the appearance of this noise.

This is where the Canon vs Sony-Nikon sensor argument arrises.
One company's sensors have more pattern noise at low iso shadows than the other company's, thus limiting the ability to boost shadows and achieve an effective HDR image from one exposure.

EXACTLY. Thank you for concisely stating the reason why some of us care about banding in so-called 'useless shots taken with the lens cap on'.