Uwe Steinmuller of DOP on dynamic range and HDR

[hjulenissen]

I think it's a little unfair that HDR gets such a bad rap when people are Topazing the shit out of images and no one seems to bat an eye.  Or when guys like Dave Hill develop a processing methodology that creates anything but a realistic look and people fawn over it.  Doesn't make a lot of sense.  What it does do; however, is provide further proof that the appreciation of 'art' and what is or isn't 'art' is entirely subjective.  

For one, I would expect 'art' to be something different, or more than 'a realistic snap of a scene'. B&W? Filmgrain? non-linear response of film? Eerie long exposures of waves that does not map to anything that I can see with my bare eyes?

If anything 'non-realistic' is bad, then a lot of photography is bad. If some non-realistic photography is good, then no photography should be dismissed for being non-realistic. Perhaps for being 'to radical compared to what we are culturally used to', or 'too easy to accomplish for casual users and therefore not worthwhile', or simply 'not according to my taste'.

-h

[JimAscher]

For one, I would expect 'art' to be something different, or more than 'a realistic snap of a scene'. B&W? Filmgrain? non-linear response of film? Eerie long exposures of waves that does not map to anything that I can see with my bare eyes?

If anything 'non-realistic' is bad, then a lot of photography is bad. If some non-realistic photography is good, then no photography should be dismissed for being non-realistic. Perhaps for being 'to radical compared to what we are culturally used to', or 'too easy to accomplish for casual users and therefore not worthwhile', or simply 'not according to my taste'.

-h

You have managed to sneak in, possibly "under the radar," a profound reminder on the subject of photography as art, which is regrettably somewhat rare in this forum.  Many thanks. 

[John Camp]

For one, I would expect 'art' to be something different, or more than 'a realistic snap of a scene'. B&W? Filmgrain? non-linear response of film? Eerie long exposures of waves that does not map to anything that I can see with my bare eyes?

If anything 'non-realistic' is bad, then a lot of photography is bad. If some non-realistic photography is good, then no photography should be dismissed for being non-realistic. Perhaps for being 'to radical compared to what we are culturally used to', or 'too easy to accomplish for casual users and therefore not worthwhile', or simply 'not according to my taste'.

-h

I think you're wrong on almost all of this.

Why should art be something different than you can see with your eyes? Jeff Wall takes high-resolution, very naturalistic photos of scenes that he creates much as a movie director does, but what you get in the photo is exactly what was in front of the camera. The art is in the creation, not in what the camera does.

Nobody said everything non-realistic is bad. You've set yp and knocked down a straw man. If some non-realistic photography is good, you can still dismiss other non-realistic photography as bad, even for no other reason than it's non-realistic. The question is, does the work succeed in its own terms? If somebody says, "We used HDR to increase realism in this photo," and they didn't increase realism, then they failed in their own terms. Sometimes, that's hard to tell, but it usually isn't.

As a general proposition, I'd suggest that any sweeping statements about art, such as yours, are wrong.

[Guillermo Luijk]

A bit dissapointed to read this:

"If our cameras could capture high dynamic range scenes in a single shot we wouldn't need the techniques described in these articles."

It makes me think Steinmueller didn't really get that the point of bracketing for HDR will soon be unnecessary, and it does not participate in the definition of HDR itself. The only reason we have today for bracketing HDR scenes is that sensors are still too noisy to capture in a single shot the entire DR of many real world scenes.

But eventually (and this is already close to happen, just take a look at the amazing DR of most recent APS-C sized sensors used in the Pentax K5 and Nikon D7000), we won't need to bracket. Will that day be the end of HDR? not at all. The HDR problem will remain because HDR is about tone mapping the captured HDR onto LDR devices such as the monitor or the print.

So the Photomatix develop team can stay happy, people will go on using their software in the future. The only difference will be that today's bracketed input files will become a single input file with all the information on it. Easier for users as well: bye bye to all those alignment and ghosting issues in non-static scenes.

BTW from my experience I think Steinmueller's DR figures for the output devices are too optimistic:

"Today's Monitors: 1:300-1:1000 -> 8,2-10 stops
HDR monitors 1:30000 (watch your eyes, may get stressed) -> 14,9 stops
Printers on glossy media: about 1:200 -> 7,6 stops
Printers on matte fine art papers: below 1:100 -> 6,6 stops"

I have measured real DR in normal observation conditions (i.e. ambient lighting) and my HP LP2475W monitor yielded 6,7 stops (vs 8,2-10), and a printed copy on Fujifilm glossy paper yielded 4,3 stops (vs 6,6-7,6). I'd love to find out what an HDR monitor looks like!.

Regards

[LKaven]

A bit dissapointed to read this:

It makes me think Steinmueller didn't really get that the point of bracketing for HDR will soon be unnecessary, and it does not participate in the definition of HDR itself. The only reason we have today for bracketing HDR scenes is that sensors are still too noisy to capture in a single shot the entire DR of many real world scenes.

Surprised to read this from you!  One very good reason for bracketing exposures is because of the properties of supersampling a scene into 32-bit space.  Just the increase in fidelity on the lowest tones is worth the effort.  Certainly, the newer cameras will be quite accurate at quantizing the lowest tones in a scene into 2-3 bit quantities, but only with supersampling do you stand a chance of increasing the resolution of those tones.  Of course, if one is shooting digital as though it were slide film, this might matter less.  But to the rest of us, it matters.

On the other hand, I think you are very much right to identify tonemapping as something in its own realm, independent from HDR. 

[NikoJorj]

One very good reason for bracketing exposures is because of the properties of supersampling a scene into 32-bit space.  Just the increase in fidelity on the lowest tones is worth the effort.

Well, did you read and see this? Seems that 16bits already allow a fair amount of margin.

[RFPhotography]

BTW from my experience I think Steinmueller's DR figures for the output devices are too optimistic:

I have measured real DR in normal observation conditions (i.e. ambient lighting) and my HP LP2475W monitor yielded 6,7 stops (vs 8,2-10), and a printed copy on Fujifilm glossy paper yielded 4,3 stops (vs 6,6-7,6). I'd love to find out what an HDR monitor looks like!.

Regards

Steinmuller is using the theoretical max. based on the simple log equation converting a contrast ratio into a certain number of stops of light.  The log base 2 of 100 is 6.64.  Of 200 is 7.64.  In practical terms, the result is likely to be different.  But yes, I'd agree that his contrast ratios are too optimistic.  Paper prints don't carry nearly that much contrast.

John, you've made the same mistake as many others.  You've done it with respect to art in general as opposed to the ones who address HDR specifically.  You've imparted your objective position onto a subjective subject.  And that is what's wrong.  And that's not a subjective issue.  HJ has suggested what he 'expects' art to be.  An expectation isn't a hard and fast, objective construct.  Anything that captures or freezes a moment in time isn't realistic.  If I can't go to that place and see exactly what is in that photo or painting or movie or drawing or 3D rendering then it's not realistic.  The only true realism is what I, or anyone else, can see with my own eyes.  I can choose to believe or not the reality someone else saw and the way they present that reality to me and accept it as real but it's not truly real to me. 

[Ray]

John, you've made the same mistake as many others.  You've done it with respect to art in general as opposed to the ones who address HDR specifically.  You've imparted your objective position onto a subjective subject.  And that is what's wrong.  And that's not a subjective issue.  HJ has suggested what he 'expects' art to be.  An expectation isn't a hard and fast, objective construct.  Anything that captures or freezes a moment in time isn't realistic.  If I can't go to that place and see exactly what is in that photo or painting or movie or drawing or 3D rendering then it's not realistic.  The only true realism is what I, or anyone else, can see with my own eyes.  I can choose to believe or not the reality someone else saw and the way they present that reality to me and accept it as real but it's not truly real to me. 

I'm getting a sense this argument is becoming convoluted and confused.

The problem as I see it is both the eye and the camera have limited dynamic range. The eye has the disadvantage of a very narrow FoV (excluding peripheral vision which detects only movement), but has the advantage of an easily executed rapid change of direction of view, combined with a continuously changing aperture to accommodate changing brightness levels in whatever scene is being viewed.

Without bracketing of exposures, do we miraculously expect the camera with its fixed aperture to faithfully capture a scene from the brightest part of the sky to the darkest shadows, shadows which are in fact, from the eye's perspective, not dark at all, because the eye's aperture in a fraction of a second has changed from F8 to F2.8, as its gaze is directed at such darker areas of the scene?

It seems to me to be a tradition in photography and many styles of paintings (I'm thinking here of Caravaggio) to unnaturally darken parts of an image for artistic impact. Black shadows create a sense of  'pop'. They also have the effect of removing distracting elements in the image, similar to the effect of a shallow DoF.

If you want to create a piece of art which does not represent what the 'average' eye saw, but which represents a whole lot of cultural ideas, personal preferences and idiosyncracies, then almost anything goes, depending on which authoritative figure endorses the work.

As I see it, the purpose and goal of merging different exposures to HDR is to mimic how the eye behaves as it views a scene, in order to reproduce a composite (merged) image which includes all the detail in the scene which the eye would have witnessed.

Compressing that wide dynamic range to fit naturally on a medium such as monitor or print is the problem.

It's a problem that requires skill in image processing, as well as sophistication of software.

[Guillermo Luijk]

Surprised to read this from you!  One very good reason for bracketing exposures is because of the properties of supersampling a scene into 32-bit space.

Bracketing will always mean an advantage in minimising noise and having greater tonal richness (BTW no need of 32-bit floating point formats for that, a 16-bit integer with gamma can encode 99,99% real world HDR scenes. Try to download this TIFF file: superhdr.tif that can be pushed 12EV without noise or posterization).

What I mean is that with sensor technology becoming lower and lower in noise, the advantages of bracketing will eventually vanish compared to the advantages of not having to bracket (no misalignment, no ghosting issues in moving scenes, no unnecesary resources wasted to store and process several RAW files, no tripod needed in some cases,...). So with the new cameras coming, users will start to take a single shot in scenes that they are bracketing today, in the same way as today nobody takes the mess to bracket low or medium DR scenes (< 8-9 stops).

But that day will not mean the end of HDR techniques, they will still be necessary because...

Compressing that wide dynamic range to fit naturally on a medium such as monitor or print is the problem.

It's a problem that requires skill in image processing, as well as sophistication of software.

I would add this is a problem that will never have a 100% satisfactory solution, just different approaches closer to the ideal goal, and always subject to the user's subjective opinion.

Regards

[hjulenissen]

I think you're wrong on almost all of this.

Why should art be something different than you can see with your eyes? Jeff Wall takes high-resolution, very naturalistic photos of scenes that he creates much as a movie director does, but what you get in the photo is exactly what was in front of the camera. The art is in the creation, not in what the camera does.

I dont know him. Does his images look like they were taken at any random place at any random time, or does it look like he has carefully chose time, place and camera settings to make a visually pleasing image?

If he in any way is "putting his soul" into his image, I would say that that could detract from the realism but add to the artistic value.

BTW, do you think that art should be valued from the end-result alone or does knowledge of the process add/subtract to its value? If I show you an amazing image that blows your socks off (purely hypothetically speaking), would you be any less impressed if I told you I had made it purely in Photoshop? Or is the ideal that one should wait for weeks in a cold, deserted place waiting for "just the right light" and then capture that magic moment right before the batteries run out and being tragically eaten by a bear?

Nobody said everything non-realistic is bad. You've set yp and knocked down a straw man. If some non-realistic photography is good, you can still dismiss other non-realistic photography as bad, even for no other reason than it's non-realistic. The question is, does the work succeed in its own terms? If somebody says, "We used HDR to increase realism in this photo," and they didn't increase realism, then they failed in their own terms. Sometimes, that's hard to tell, but it usually isn't.

You are mixing arguments here. If "lack of realism" is a valid argument against some art it should be a valid argument against all art. If the critique is that it "is not suceeding in its own terms", then that it the argument that you should use.

You second statement statement seems irrelevant to what I said.

-h

[JimAscher]

This is a great, and quite often profound, discussion, about art (among other things).  Keep it coming.

[hjulenissen]

I would add this is a problem that will never have a 100% satisfactory solution, just different approaches closer to the ideal goal, and always subject to the user's subjective opinion.

Just like camera sensor DR is being improved, I believe that display DR is being worked on. I dont know about paper.

Using clever zoned backlighting, the impression of DR of LCD panels can be greatly improved. It might not matter if two neighbor pixels are 1000:1 or 100000:1 apart in brightness, but it may matter if larger areas are.

I really like the discussion on how our eye "scans" the scene, and adaptively adjust gain along the way to form a mental image that contains details in both darker and brighter parts. Reproducing the scene accurately would of course suffice, but what is more "natural" when you are limited to 100:1 or 1000:1 contrast is not evident. Perhaps we are only culturally trained into thinking that clipped whites and blacks (and occasionally some compression in the middle) is the most natural way of solving this problem, while some other society concievably could have convinced themselves into thinking that heavy tone-mapping was most natural. If digital processing was invented before film (or canvas), things could have turned out very differently.

-h

[LKaven]

Bracketing will always mean an advantage in minimising noise and having greater tonal richness (BTW no need of 32-bit floating point formats for that, a 16-bit integer with gamma can encode 99,99% real world HDR scenes. Try to download this TIFF file: superhdr.tif that can be pushed 12EV without noise or posterization).

The benefits of encoding in a 32-bit floating point space might not be so keenly felt in the higher tones.  But consider the 2-3 bit quantization for the lower tones in a single shot capture.  The color palette collapses into dither as you go lower.  But if you bracket and move to HDR space, you can expand that palette for purposes of post processing, and then decide where you want to map it on the tonal scale without significant loss of fidelity.

I'm not sure everyone realizes that the HDR technique involves a move into the space of absolute magnitudes, and away from relative white-black point of a single capture.  This is a conceptual shift.  I think some here are carrying over the assumption that HDR is just another tool for doing LDR, but the conceptual shift is more significant.

[LKaven]

Perhaps we are only culturally trained into thinking that clipped whites and blacks (and occasionally some compression in the middle) is the most natural way of solving this problem, while some other society concievably could have convinced themselves into thinking that heavy tone-mapping was most natural. If digital processing was invented before film (or canvas), things could have turned out very differently.

Right on the mark!  If not "heavy" tonemapping, then "sophisticated" tonemapping.

[Guillermo Luijk]

Perhaps we are only culturally trained into thinking that clipped whites and blacks (and occasionally some compression in the middle) is the most natural way of solving this problem, while some other society concievably could have convinced themselves into thinking that heavy tone-mapping was most natural. If digital processing was invented before film (or canvas), things could have turned out very differently.

This is an interesting point. I would never admit that a heavily tone mapped HDR image is closer to what my eyes perceive than a softly tone mapped image, lacking in local contrast and obtained with a simple S-shaped tonal curve, maybe with some black/highlight clipping. But perhaps this is because I have spent some 25 years looking at printed images representing HDR scenes that unavoidably became compressed to those poor 4 stops the print offers, and I have assimilated that to be the most natural look.

What is true and will always be, is that any output device with effective DR capabilities below the DR of the original scene, will never manage to make use perceive exactly what we perceived when looking at the real scene. We can only try to mimic what we perceived.

Regards

[Guillermo Luijk]

The benefits of encoding in a 32-bit floating point space might not be so keenly felt in the higher tones.  But consider the 2-3 bit quantization for the lower tones in a single shot capture.  The color palette collapses into dither as you go lower.  But if you bracket and move to HDR space, you can expand that palette for purposes of post processing, and then decide where you want to map it on the tonal scale without significant loss of fidelity.

There is absolutely no need to expand anything beyond the scene's DR. So as long as your camera can capture the DR of the scene in a single shot (and this just means acceptable number of levels and acceptable SNR for the deep shadows), there is no benefit for bracketing.

Consider the following example: a typical {0, +2, +4} bracketing from a 12-bit & 8-stops DR camera (Canon 350D), will be 100% equivalent to a single shot from a 16-bit & 12-stops DR camera (a FF camera with the same photosites as the Pentax K5 and a 16 bits ADC).

As long as sensors become less noisy, bracketing will become useless. This is a fact.

[feppe]

Perhaps we are only culturally trained into thinking that clipped whites and blacks (and occasionally some compression in the middle) is the most natural way of solving this problem, while some other society concievably could have convinced themselves into thinking that heavy tone-mapping was most natural. If digital processing was invented before film (or canvas), things could have turned out very differently.

While that's an intriguing proposition, it amounts to not much more than mental masturbation. Keke Rosberg (a Finnish Formula 1 race driver) had a great quote regarding such things: "if mother had balls she'd be the dad."

[hjulenissen]

While that's an intriguing proposition, it amounts to not much more than mental masturbation.

For something to be mental masturbation, I would have to do it in solitude, and not on a public forum, I think? :-)

My point was that "HDR"*) is controversial among photographers. Some think that it is the best thing since sliced bread, while others think that it is horrible. Some of the last cathegory will claim that HDR looks unrealistic (implicitly saying that regular LDR looks realistic). I dont think anyone can argue from mathematics that one or the other is more similar to the original - HDR preserves some aspects of the true scene, while regular LDR preserve other aspects of the true scene. So we are left with arguing what "looks more similar to me". We cannot throw out 100 years of cultural baggage instantly, but culture may change in years (while the human visual system may need 100 generations to change significantly). Therefore, the answer to my "mental masturbation" could tell us if HDR may be the accepted norm in 10 or 20 years, or if it will be a quickly passing fad.

People were sceptical towards stereo sound as well - and early recordings gave good reason. When creative people get a new tool, they tend to use it everywhere. Now, stereo is the norm, and recording technique have matured to use it with sense (or our culture have adopted to its sound, who knows). The excessive use of phasers and flangers on vocals and drums have all but disappeared, though (some might think that is a good thing) - either we could not get used to them, or producers could not create enough variation within that sound to still sound "fresh".

*)I adopt the imprecise convention of using the term "HDR" to describe the joint process of HDR capture (usually through multiple exposures) and tone-mapping to LDR

[BJL]

Just like camera sensor DR is being improved, I believe that display DR is being worked on. I dont know about paper.

About paper, and any display relying on reflected light rather than transmitted or emitted, I am fairly sure that the brightness range displayed will stay well below what even the humblest SLR photosites are capable of recording. For one thing, the lowest reflectivity of any natural substance is about 2%, so the range from that to perfect 100% reflectivity is only about 50:1, or under 6 stops. Short of exotica like printing black with carbon fiber nanotube material (as in NASA's new super-black coating for flare control in telescope lenses), even 8 stops is out of reach of prints.

But this is maybe just one more reason that many of us now prefer the display screen to the print for more realistic, vibrant reproduction of what we saw when we were taking the photo. (And why many prefer slides to prints -- RIP Kodachrome, December 30, 2010.) I will admit that already, my favorite way to view my images is the big, bright, high brightness range 19"x13" images of a 23" diagonal computer screen, despite the relatively low resolution compared to what my files contain and prints can reveal. For one thing, where I would move closer to examine details within a large, sharp print, I can instead stay at a comfortable viewing distance and pan and zoom on the screen, so it is mostly enough for the display to have enough resolution for a "normal" viewing distance, roughly equal to image diagonal.

So maybe (at least to avoid the need for display pan&zoom), I should worry less about advancing the IQ of my "capture devices" and think more about when, if, and how my "display devices" will catch up with what my capture is already giving, in the sense of simultaneously displaying all the resolution and all the brightness range. Maybe something like the 326ppi of the new iPhone/iPod Touch "retina displays" scaled up to a 23" (or 19"x13" or A3) screen, for about 25MP. Or at least a more modest "photo quality" 200ppi, and so about 10MP.

[JeffKohn]

My point was that "HDR"*) is controversial among photographers. Some think that it is the best thing since sliced bread, while others think that it is horrible. Some of the last cathegory will claim that HDR looks unrealistic (implicitly saying that regular LDR looks realistic). I dont think anyone can argue from mathematics that one or the other is more similar to the original - HDR preserves some aspects of the true scene, while regular LDR preserve other aspects of the true scene. So we are left with arguing what "looks more similar to me". We cannot throw out 100 years of cultural baggage instantly, but culture may change in years (while the human visual system may need 100 generations to change significantly). Therefore, the answer to my "mental masturbation" could tell us if HDR may be the accepted norm in 10 or 20 years, or if it will be a quickly passing fad.

Actually I think you can argue mathematically that the heavily stylized HDR look tends to be less realistic. It's pretty common to have tonal inversions in these types of images, where for instance the shadowed foreground is actually brighter than the daytime sky, just to name one very common example. So I don't really think you can argue that folks think this stuff looks unnatural just because film came first.  Maybe if the real world looked like the one in Avatar this argument might hold some water...

If some people like the stylized HDR look with aggressive "detail enhancement" that's fine. Different people have different tastes; and when it comes to art anything goes, so I certainly don't think that a naturalistic approach is the only valid one. I can appreciate truly well-done stylized HDR, even if it's not to my personal taste. The problem is, it's extremely rare. The vast majority of stylized HDR imagery is full of ugly artifacts that I just can't see past, and it boggles my mind that so many people don't seem to mind the ugly halos, color shifts, etc. Hopefully over time the tools will get better and this will improve; but right now I would say that the "bad" HDR outweighs the good by at least 10:1. So for a lot of people, this pretty much spoils the whole genre.