Can HDR be done with just one exposure?

[MikeFairbanks]

I heard that if you shoot raw you can put it into HDR (with photomatix) and get a good result.

True?

[fly my pretties]

Yes, you can get a fairly decent result with a single RAW.

Here is one I did a few weeks ago:

http://i28.tinypic.com/149s8xv.jpg

[fly my pretties]

One more from the same night:

http://i32.tinypic.com/nbqe0g.jpg

[kirkt]

The short answer to the question posed in the subject line of the OP is "No." A single RAW does not provide the dynamic range to be considered High Dynamic Range. You can, however, run a single RAW image through the tonemappers in HDR apps like Photomatix, etc. and get pleasing results. A single RAW, tonemapped thusly, is sometimes called a "pseudo-HDR." The term HDR has, unfortunately, become synonymous with the "look" that the contrasty local tonemapping operators give, as opposed to the generation of a true high dynamic range dataset made from multiple exposures of a scene (a scene-referred luminance map).

Kirk

[ArcticShooter]

Great answer Kirk!

[ddong]

The answer is YES, even without using PhotoMatrix. However, you have to use RAW file for better final result.

You can always increase or decrease the image exposure by 1 or 2, doesn't matter if it's JEPG or RAW, then combine them together by using image processing software.

[fly my pretties]

The short answer to the question posed in the subject line of the OP is "No." A single RAW does not provide the dynamic range to be considered High Dynamic Range. You can, however, run a single RAW image through the tonemappers in HDR apps like Photomatix, etc. and get pleasing results. A single RAW, tonemapped thusly, is sometimes called a "pseudo-HDR." The term HDR has, unfortunately, become synonymous with the "look" that the contrasty local tonemapping operators give, as opposed to the generation of a true high dynamic range dataset made from multiple exposures of a scene (a scene-referred luminance map).

Kirk

In other words, the answer is yes, just don't call it HDR because you'll displease the purists. The rest of the world will still call it HDR.

[kirkt]

Well, the subject of the post was "Can HDR be done with just one exposure?" The answer is "No" because HDR means "High Dynamic Range" - a process which involves extending the dynamic range of the typical camera sensor by combining multiple exposures into a single dataset - the technique allows capture of scene dynamic ranges that exceed the dynamic range of the camera's sensor. Because the data may have such a large dynamic range, they are typically not displayable in their native form on monitors and printable on paper, etc. THerefore, these datasets need to be tonemapped to produce displayable/printable images. However, because they contain the entire scene-referred luminance data, they can be used to generate lighting for CG environments, compositing etc., and can also be manipulated to produce different "exposures" of the same scene through tonemapping. This is what "HDR" means; however, with the increase in the popularity of using HDR datasets in photography to capture a scene that would otherwise not be possible in a single exposure or without extra lighting, etc., the term "HDR" has become increasingly used to describe the *effect* of the tonemapping operators used to make true HDR datasets viewable on LDR devices like monitors and printers.

There is a difference, but the "rest of the world" who do not understand this may not care. Because this is the "HDR" part of the forum, I assumed you were interested in HDR and understanding the difference between what HDR means versus what people may mean when they use the term HDR.

If you mean, "can I run any old image through a tonemapper and get results" - the answer is "Yes". However, the result will not be the tonemapped version of an HDR dataset, it will be a contrasty version of the LDR data that you fed the tonemapper. The image, from the moment you captured it until the moment you ran it through the tonemapper, was never HDR. The difference is not one of sematics that may upset the purists. That being said, I recognize that the term "HDR" has come to mean the "look" that tonemapping a true HDR dataset may give. As ddong mentioned above - you can take a single image and process it multiple times to yield and image "sequence" but it still won;t be HDR because you are not generating any new data when you do so. THis approach is sometimes called "pseudo-HDR". With a RAW image, a 10-12 bit image, you typically have more data to work with than a single 8-bit image.

I apologize if I offended your sensibilities.

Kirk

[eviltech]

you go KIrk, well put.

[MikeFairbanks]

That was a great response, Kirk. Thanks.


I'm very new to HDR, and am really enjoying learning. My results so far are not good, but they're getting better.

[JJD.Photography]

What is the best way to create the other 2 images when working off a single Raw using Photoshop CS3, LR, or Photomatix?

I have tried using the single exposure through Photomatix, but getting bad results (lot of noise).

[digirebelva]

If I am only using 1 shot I will run it through the raw converter of choice..(zoombrowserEX for me) and make 3 tiffs from it..A normal exposure & 1 each of +2 & -2 and then combine the tiffs in photomatix removing the option of aligning the photos in photomatix.
Below I one I just did last week using that process

http://farm3.static.flickr.com/2446/3720675014_08e112064e_b.jpg

[Gnarlyographer]

In my own words I'd like to say HDR is from many differant exposures and looks amazing but even without 3 or more shots you can get the HDR look, But also what a few freinds and myself call single shot HDRs as FDR "full dynamic range" no extra exposures need to be made nor even post prossesing with good light and the right choice of camera settings you can make an image that has neither blown highlights or covered shadows.( dodge and burn a bit and crank some midtone contrast and ppl will believe this is HDR)

Now the short answer.
YES, you can make single shots to have a "HDR look".

(I THINK "KIRKT" SAYS IT BEST AND DESCRIBES IT IN DEPTH)

[_GUI_]

In other words, the answer is yes, just don't call it HDR because you'll displease the purists. The rest of the world will still call it HDR.

It is not a question of being purist or not, it is a question of DYNAMIC RANGE. Do you have a clear understanding of what DR means? I will answer for you: you don't.

I hope the very well informed answers from kirkt don't get blinded by these opinions

Regards.

[kirkt]

That was a great response, Kirk. Thanks.


I'm very new to HDR, and am really enjoying learning. My results so far are not good, but they're getting better.


Cool. This site I find my personal favorite for good general information, interesting and informative news, and a ton of tips and tricks:

http://www.hdrlabs.com/news/index.php

It is a mix of HDR purely for photography, HDR for CG (the author Christian Bloch is a VFX artist) and pano/QTVR HDR. If the CG stuff is not for you, no worries. The approach to generating good quality HDR datasets is the same. Consider purchasing Christian's book - it is an excellent reference for HDR, assembling panos and a ton of other HDR and tonemapping techniques and, should you get into it, you will find almost every HDR app out there gives significant discounts to people who have purchased the book via the website (you usually have to answer a question like "what is the last word on page 145?" to get the discount). Read the forum too, Christian is a funny dude.

Good luck, it is a pretty cool technique and gives flexibility with lighting in tough situations. Once you get the concepts under your belt you will probably better understand some of the jargon I was throwing around in my long-winded explanation earlier in the thread.

Have fun!

Kirk

[fly my pretties]

It is not a question of being purist or not, it is a question of DYNAMIC RANGE. Do you have a clear understanding of what DR means? I will answer for you: you don't.


You don't have to answer for me, because a) I'm not an idiot, and b) I know exactly what dynamic range is.

I would know what dynamic range was even if I wasn't a photographer, since the noun 'range' is defined by the adjective 'dynamic'. Those two words tell me, that 'dynamic range' refers to the distance between the largest and smallest physical values of changeable quantity.

By the way, kirkt, when you describe it like that you make a lot of sense.My real point, which I should've explained in detail is that if you have a properly exposed picture, and you create two more exposures 2 stops apart from that single image, you are going to end up with the same visible luminance range than you would from three seperately exposed images. Granted, I gues you could argue that each of those three images than be adjusted, but nobody does that. I think 99.9% of people would be unable to tell the differnce between a tonemapped image from a single properly exposed shot, and a "true" HDR image from three exposures.

(I say three, because that keeps the discussion fair)

[René Damkot]

I'd like to remind all that discussion is good, name calling isn't...

[CJinAustin]

I heard that if you shoot raw you can put it into HDR (with photomatix) and get a good result.

True?

If you create two more files from the original raw, one over a stop and one under a stop, you can then put those files through photomatix and it will work ok. It will certainly have a broader dynamic range than just developing the original file by itself. But you will get better results if you can take multiple exposures in camera.

[Gibbo]

http://farm3.static.flickr.com/2446/3720675014_08e112064e_b.jpg

That is a brilliant example to answer "Yes you can use one" ;)

Here's my contribution.. A quick play-around. I was just experimenting with using one RAW.

Whilst walking through Munich, Germany

http://farm4.static.flickr.com/3313/3328958500_7a4edd00fa_b.jpg

[District_History_Fan]

In other words, the answer is yes, just don't call it HDR because you'll displease the purists. The rest of the world will still call it HDR.

The rest of the world should be calling it a tonemapped raw file... ;)

[Scottes]

That is a brilliant example to answer "Yes you can use one" ;)
Those are wonderful tonemapped images, but they are not High Dynamic Range images since it has no more dynamic range than the original, single image had.

HDR is not tonemapping, and tone mapping is not HDR. Tone mapping is one way to present an HDR image using an LDR viewing medium.

[_GUI_]

if you have a properly exposed picture, and you create two more exposures 2 stops apart from that single image, you are going to end up with the same visible luminance range than you would from three seperately exposed images.

Dynamic range is not just a matter of luminance, but of information too. With just one shot you cannot capture the same amount of information as with 3 (or 2, or more than 3) captures with different exposure.

In fact with just one shot in a high dynamic range scene you will get:
1. Blown information where the 3 shots preserved the highlights
2. So much noise in the shadows that the textures become unrecognisable
3. Both 1 and 2

Said that, to capture a high dynamic range one shot is not enough.

"A properly exposed picture" is a funny phrase if we are talking about high dynamic range. You cannot capture a 14EV scene with just one properly exposed shot because the term 'properly exposed' for such scenes doesn't make sense since your camera cannot go beyond 8-9 EV. This is what high dynamic range is about, and not about tone mapping which is your only concern.

This is a 12 f-stops high dynamic range scene:

http://www.guillermoluijk.com/tutorial/hdr/resultadolite6.jpg

If you shoot to keep detail in the highlights, the shadows become noisy and textures are lost (RAW1).
If you expose to have detail in the shadows, then the highlights get blown (RAW2).
Conclusion: with just one single shot you cannot capture all its high dynamic range.
But with 2 shots of carefully chosen exposure values, the entire dynamic range is captured (RAW1+RAW2=RAW VIRTUAL):

http://www.guillermoluijk.com/article/virtualraw/comp.jpg

These kind of high dynamic range scenes are the ones where talking about HDR makes sense, and these are the kind of scenes where one shot is not enough. Talking about HDR in a low dynamic range scene that can be captured by the camera in a single shot is nonsense. In those situations people say HDR when they really mean artificial tone mapping.

Regards.

[kirkt]

To understand the idea behind HDR imaging, do yourself a favor (I use the royal "yourself", meaning "anyone who cares to do the experiment") and perform the following experiment. It requires the use of a tripod for image alignment. Shoot in RAW for this particular experiment, as will be described below.

1) The setting for your shot will be similar to _GUI_'s example above - that is, an interior, during the day, with a large exterior portal to natural sunlit exterior scenery. This is a typical example of a high dynamic range scene, where the range of scene luminance may go as high as 12-16 EV, definitely higher than the ability of your typical dSLR's sensor.

2) The shot should include deep shadows in the interior space and the brightest highlight areas of the exterior scene. Frame the shot and take some test images to determine the range of exposures required to capture the deepest shadow and brightest highlight data properly.

3) The technique for capturing an HDR image sequence is discussed all over the net and will not be explained here in depth. Everyone has their own preference for doing so, but I will give you a good basic technique here. Put your camera in "Manual" mode. Set the WB to Sunlight. Set the aperture to something reasonable like f/8 and the ISO to something reasonable as well, like 200. If your camera has a self timer, use it. If it has the ability to use mirror lock up, use that too. I shoot with a 5D, using mirror lock up. When I use the timer with mirror lock up, the timer interval is 2 seconds, nice and short.

3) To change the exposure in your image sequence, you are not going to mess with any setting other than shutter speed. Set the shutter speed to the slowest speed required to expose the deepest shadows properly. Let's say for our example here that that shutter speed is 4 seconds. Snap your image. Now, for the experiment we are going to collect a bunch of data - some of it may be redundant, but that is okay. So, we are going to collect images at 1EV increments. Therefore, the series of images you shoot will be at shutter speeds equivalent to 1EV bumps in exposure. Let's say the highlights in the scene are properly exposed at 1/2000 sec (very bright outside!). So our exposure sequence would look like:

4 sec, 2 sec, 1 sec, 1/2 sec, 1/4 , 1/8 , 1/15, 1/30, 1/60, 1/125, 1/250, 1/500, 1/1000, 1/2000

for a total of 14 images.

You may not need to shoot at 1EV increments, but that is for another experiment. Here we are collecting a lot of data so we can understand what is going on. Also, your scene may vary and require a smaller range of exposures, so shooting at 1EV increments may only require 7 or 8 images to be collected.

4) Now you have all of the RAW data. The experiment will involve making a true HDR dataset from the collected RAW images in the exposure sequence, versus taking the "best" single RAW and creating multiple versions of it by bumping the exposure slider in ACR. Do that experiment and you will quickly realize that the generation of multiple "exposures" from a single RAW generates no new data - so, if the scene contains a range of luminance that is beyond that of your sensor, you cannot create that uncaptured data by sliding a slider in ACR. This is especially noticeable in the shadows, where fewer levels of data exist to begin with (see _GUI_'s example above).

If the scene itself does not contain a range of luminance values (a dynamic range) that exceeds the range of your camera's sensor, then a properly exposed single shot will capture the entire range of luminance for that scene (like an overcast, diffusely lit day). This kind of scene is not a "high dynamic range" scene. You can, however, still run that RAW file through a tonemapper and get a contrasty version of the image as output.

It should be noted that "tonemapping" in and of itself is not unique to presenting HDR data. When you use "Levels" or "Curves" in Photoshop or similar apps, you are tonemapping - that is you are remapping the relationship between the original tonal values in the image data to new tonal values. The tricky bit with tonemapping HDR data is that you must map the potentially very large range of HDR tonal data into a relatively small range of values that are displayable or printable. These techniques often use local or global contrast operators that, for each pixel, examine the tonal values of the surrounding pixels to come up with the new value for that pixel. These specialized operators give the tonemapped version of the data the "look" that people identify as "HDRish".

Do the experiment and convince yourself.

Have fun!

Kirk

PS - regarding the interior-exterior image typical of HDR, there are other ways to "tonemap" and produce an image that is similar to what your eye perceives when you are standing in the scene. This is often done by lighting the interior scene so that its brightness values fall closer in exposure value to the exterior brightness - thus, the RANGE of brightness in the scene is compressed and can fit entirely within the sensor's range. Clever, eh? Same idea, different approach. This approach requires a completely different mindset though, as interior lighting that does not disturb the natural look of ambient light is a tricky art, etc. The same idea applies to using fill flash outside in harsh sunlight - you are trying to bring the brightness of the backlit person's face closer to the brightness of the surroundings so that the range of brightness in the scene can be balanced and captured in a single shot.

[Gibbo]

Those are wonderful tonemapped images, but they are not High Dynamic Range images since it has no more dynamic range than the original, single image had.

HDR is not tonemapping, and tone mapping is not HDR. Tone mapping is one way to present an HDR image using an LDR viewing medium.

Ermm.. no, it does have more dynamic range than the original.

I maybe wrong, care to explain why?

[Scottes]

You captured a single exposure with a dynamic range limited by the capabilities of your camera.

If you had areas that were blown out during this exposure you failed to capture the details in this area. If some parts were so dark that they registered as black on the sensor you failed to capture the details in this area. If you had neither black nor blowouts then you failed to capture the dynamic range that the camera was capable of capturing.

Your tonemapped image does NOT have more dynamic range than the single exposure.

Making some pixels brighter does not increase the dynamic range of an image.

[Scottes]

An example, in reverse.


The original capture:

http://www.itsanadventure.com/postimages/GBH_59749-org.jpg

The results after post-processing:

http://www.itsanadventure.com/postimages/GBH_59749.jpg

I used post-processing the manipulate the colors to INCREASE the range of shades displayed.

I did NOT change the dynamic range of the original image.

Tone mapping is a post-process method of manipulating the shades in order to DECREASE the shades of an image so that the image can be displayed in a low-dynamic medium such as a monitor.

Tone mapping does NOT change the dynamic range of the original image.

[fly my pretties]

kirkt, thanks for the indepth explanation on HDR. After reading what you've written, I understand that this is not merely an argument of sementics, and concede that I was completely in the wrong with my initial assessment of HDR.

[kirkt]

@fly my pretties - no worries. That's why I took the time to write it up. It can be confusing, especially when the terminology has morphed into more of a catch-all phrase.

Have fun!

Kirk

[Harreh]

Yes, you can get a fairly decent result with a single RAW.

Here is one I did a few weeks ago:

http://i28.tinypic.com/149s8xv.jpg

How very mysterious, I love it!

[kirkt]

http://www.hdrlabs.com/news/index.php?id=3090771698108409072

Has a nice discussion of the "controversy" of tonemapping and the HDR look. Please take a look at some of the links that Christian includes. As he says, HDR is growing up. It's all good.

Kirk

[polarbare]

A couple of great posts by kirkt and Scottes in this thread.

As an aside, the Boston tonemapped images I posted a few threads down were a single exposure loaded directly into Photomatix. The newer version doesn't need you to created a -2/+2 (for example) from the RAW file.

[playdoh]

Using Photomatix, would I be better off taking three raw photos at different exposures or will I get the same results with just one and let the software do the work?

Also, is it better to go -1, 0, +1 or -2, 0, +2.

Thanks so much!

[Scottes]

Shooting -1, 0, +1 or -2, 0, +2 are both the same thing, really. The other option is 0, -1, +1 in most Canon cameras. This is crazy, in my opinion. When I flip through and view the images I like to see them get progressively brighter or progressively darker. I don't want to see Middle, Dark, Bright - that's just some crazy random order.




In my opinion, more shots to cover the same EV range must produce a better final result since there's more information to work with. But this is just my theory...

If you take 3 shots 2 stops apart, versus 5 shots 1 stop apart, you cover the same dynamic range. Obviously, 5 shots gives you more information - 66% more.

So it seems obvious to me that the 66% more info in a 5-shot sequence would yield a better final result.

That's for the pixel-peepers out there, and I am one of them. I believe the above.


But to play devil's advocate....

The CPU number-crunching that the HDR software will do is much higher. Not just 66% longer, but it's probably closer to 3 times longer. (Just a wild ass guess - remember that I'm playing devil's advocate here).

The current software, though very good, probably won't take this extra data and really use it to it's fullest. So 66% more info may yield an image that is 10% or maybe 20% better. That's not really all that much.

You probably won't be able to tell the difference between the two shots. There are millions of possible colors in both shots, and the human eye can only discern 10,000 variations in a single scene (if that).


Now, to play another form of devil's advocate, let's take this thought to an extreme...


We currently think about doing HDR set 2 stops apart, because that's what our cameras do for automatic exposure bracketing (most Canons, at least).

What if our cameras could do exposure bracketing set at 3 stops apart? 3 shots at 3 stops apart would yield a wider dynamic range. That's good, right?

No, because you're spreading the same amount of information - 3 images - across a wider dynamic range. This is equivalent to upsizing/uprezzing an image to make a bigger print. It's not necessarily better, just bigger, and if you go too far the print start to get the jaggies and doesn't look as sharp.


Take this thought to a further extreme - why not just shoot on manual and take each shot 5 stops apart? Hell, 3 shots would cover a much wider dynamic range!

Yep, and the results would probably look like crapola, like upsizing a 4x6 to 11x17.



So, in the end, it's my theory that more shots set closer together will produce a final better shot. If I had the time, patience, CPU, and disk space then I would probably shoot 17 shots 1/3-stop apart, just to get the maximum amount of data into the HDR software.

But I'm a pixel-peeper, so I think that way.

I will say that I shoot a very large number of shots 1 stop apart. I may not need 9 shots to cover the scene's EV range, but at least I'll have them. And I can't always go back and shoot a scene again, so I'm going to over-shoot and save the RAW images.

Who knows, maybe in 5 years the HDR software will evolve to the point where 9 shots set 1 stop apart produces fantastic images every time. I'll have the RAW files ready for that program.


I don't know right now if shots taken 1 stop apart is better then 2 stops.

But I will continue to shoot 9 shots set 1 stop apart because I can't go back in time.

[MikeFairbanks]

This thread is still going, eh?

I guess I opened a can of worms.

Here's my latest, taken from about six shots. It's my best so far, but nothing compared to what some of you guys are doing.

http://i139.photobucket.com/albums/q295/thepirannah/ADockonaLake4.jpg

[polarbare]

Folks should also keep in mind the whole -X, 0, +X bracketing is just to make things quick and easy. People like Ben Wilmore use a much better technique (IMO) of capturing your highlight detail in the first frame, and go for longer exposures 1 stop intervals for however many exposures are needed to get all the detail you need whether it's 3,4,5,.. You'll be very happy with the results I think.

[CyberDyneSystems]

Totally agree with Kirk on this.

You CAN apply the technique using a single RAW file, and multiple conversions, as mentioned.
Heck if you were silly you could apply the technique to a few versions of the same jpeg,..

Neither would be HDR though.

The Whole Point of HDR is to get an image that shows more dynamic range than your sensor can capture.

You can mess with EC in a RAW converter, but that is not getting any more DR out of your camera than it was already capable of grabbing.

It ain't HDR without multiple exposures.

[_GUI_]

If you take 3 shots 2 stops apart, versus 5 shots 1 stop apart, you cover the same dynamic range. Obviously, 5 shots gives you more information - 66% more.
(...)
The current software, though very good, probably won't take this extra data and really use it to it's fullest. So 66% more info may yield an image that is 10% or maybe 20% better. That's not really all that much.
(...)
We currently think about doing HDR set 2 stops apart, because that's what our cameras do for automatic exposure bracketing (most Canons, at least).

What if our cameras could do exposure bracketing set at 3 stops apart? 3 shots at 3 stops apart would yield a wider dynamic range. That's good, right?
(...)
it's my theory that more shots set closer together will produce a final better shot.

Hi Scottes, 5 shots 1 stop apart don't provide 66% more information than 3 shots 2 stops apart because much of the existing information can be found in more than one of the RAW files, with nearly the same quality. Does making a copy of a given RAW file double the amount of information?.

There will be some slight improvement in visible noise in some areas of the image by shooting at 1EV intervals vs 2EV intervals, but that improvement will be negligible (the reason for this is not the HDR software not being optimum, but the way a digital sensor works). Even going from 3 shots 2EV apart to just 2 shots 4EV apart doesn't mean a big loss in quality in many cases (see sample below).

As long as you get a noise free image with the highlights information preserved, the work was done, no matter how many shots were done or how many EV apart they were. And to achieve this, 3 shots 2EV apart are 99% as good as 5 shots 1EV apart.

Think that a camera can capture very acceptably free of noise up to aprox. 6 f-stops. When performing the exposure blending step in the HDR process, only the best exposed areas of each shot will be chosen to participate in the final image. That means shooting at 2EV intervals will ideally force to use the 2 higher f-stops of the camera's dynamic range if possible, which have a very good quality on any camera.
Setting your shots 1EV apart will make us use just the highest f-stop of the sensor range whenever possible, which is even better than using both the 1st and the 2nd, but the improvement is minor.

In addition to that, there are real reasons for keeping the number of shots to a minimum:

Extra shooting time, storaging space (both in cards and HD), transferring time and processing time needed
Loss of sharpness since the final image will be more progressively blended. Unless your shots are milimetrically aligned (impossible to achieve if you touch the camera between shots), there will remain some misalignment even if an auto-align feature is used because misalignment doesn't happen in a rounded amount of pixels
Ghosting more likely to happen if there are moving elements in the scene for the same reason as 2, since there are more shots to progressively contribute to ghost the moving area


Regarding the EV gap to use, the {-2,0,+2} scheme, ensuring the '-2' shot is an ETTR (Exposed to the Right) shot preserving the highlights, can suffice for most situations. Sometimes I miss not to be able to do {-3,0,+3} for higher DR situations though.

It would be very desirable that camera manufacturers provide a highly configurable AEB. It would be nice to be able to define the number of shots, gap between them, or even any arbitrary shooting sequence (E.g. being able to tell the camera: shoot 0,+2,+5). It is ridiculous that the last Canon models remain stuck at the obsolete {-2,0,+2} scheme, I wonder if those jap engineers ever take a look at the forums.


A couple of examples to demonstrate the theory:

A scene of 12 f-stops of real dynamic range was entirely captured with just 2 shots 4EV apart: ZERO NOISE VIRTUAL RAW (http://www.guillermoluijk.com/article/virtualraw/index_en.htm). The improvement of doing more shots would have been negligible.
The info was summarized into an HDR RAW file for download here: rawvirtual.dng (http://www.guillermoluijk.com/download/superhdr.tif), which is noise free in the entire DR of the scene and preserves the highlights unclipped.

A scene of >16 f-stops of real dynamic range was captured with 5 shots 3EV apart: TIFF WITH MORE THAN 16 F-STOPS OF DYNAMIC RANGE (http://www.guillermoluijk.com/article/superhdr/index.htm).
The info was summarized into a TIFF file for download here: superhdr.tif (http://www.guillermoluijk.com/download/superhdr.tif), which can be overexposed by up to 12 stops (curves to do so are included) without any noise or posterization appearing.

NOTE: both files, the virtual DNG and the TIFF, just contain the entire dynamic range demonstrating no more shots were needed. They have not been tone mapped yet which is another step in the process, not linked with the questions about the needed amount of shots and EV gaps under discusion.

Regards

[MikeFairbanks]

It would be very desirable that camera manufacturers provide a highly configurable AEB. It would be nice to be able to define the number of shots, gap between them, or even any arbitrary shooting sequence (E.g. being able to tell the camera: shoot 0,+2,+5). It is ridiculous that the last Canon models remain stuck at the obsolete {-2,0,+2} scheme, I wonder if those jap engineers ever take a look at the forums.




Dude!

Nevertheless, if cameras were programmed to do HDR, everyone would figure it out.

I'd love to set up my XS to take five or six exposures by itself. Set the two-second timer, hit the button, and it takes all the shots for me at each programmed exposure interval.


I think one way to probably get some hands-off technique would be to hook up to a laptop and shoot that way. You can control the camera via the laptop., including exposure, shutter, etc.

[kirkt]

Nevertheless, if cameras were programmed to do HDR, everyone would figure it out.



New Pentax K-7: The first HDR camera?
from HDRLabs

http://www.hdrlabs.com/news/index.php?id=67914163388355302

[Scottes]

Hi Scottes, 5 shots 1 stop apart don't provide 66% more information than 3 shots 2 stops apart because much of the existing information can be found in more than one of the RAW files, with nearly the same quality. Does making a copy of a given RAW file double the amount of information?.
OK, I'm not going to doubt you, Guillermo - I know better than that. But I have to think that more shots 1 stop apart *has* to contain more info. I'll provide my thinking, and you tell me where I'm wrong.

This theory is from my understanding of how the camera captures light. A 12-bit camera records 4096 discrete tonal levels. If the camera can captures a dynamic range of 9 EV, then the brightest "stop" in the image would contain about half - 2048 - discrete tonal values. The next darkest stop contains one-half of that, only 1024 tonal values. The third darkest contains half again, or 512. All the way down to the darkest stop, which contains only 16 discrete tonal values.

(From "Expose Right" http://www.luminous-landscape.com/tutorials/expose-right.shtml - and Norman Koren http://www.normankoren.com/digital_tonality.html - and several other places.)

For simplicity, let's pretend the camera covers 5 stops. The chart below lists the number of discrete tonal values possible in each stop.
Brightest Stop - 2048 max possible tonal values
1 stop darker - 1024
Median - 512
1 stop darker - 256
Darkest Stop - 128Each pixel in the brightest stop has a possibility of 2048 discrete values. The next darkest stop has 1024 possible values.

...only the best exposed areas of each shot will be chosen to participate in the final image.
So why would you not want to get the most data - 2048 possible discrete values - into every "stop" in the final HDR? For simplicity's sake - which is often how my mind works - I'm thinking in stops, so I view "the best exposed area" as a 1-stop chunk of data.

This chart totals the maximum number of possible tonal values for each stop in a sequence of 3 shots taken 2 stops apart ("3@1").
EV
+4 2048 = 2048 max possible tonal values
+3 1024 = 1024
+2 2048 512 = 2048
+1 1024 256 = 1024
0 2048 512 128 = 2048
-1 1024 256 = 1024
-2 512 128 = 512
-3 256 = 256
-4 128 = 128This chart totals the maximum number of possible tonal values for each stop in a sequence of 5 shots taken 1 stop apart ("5@1").
EV
+4 2048 = 2048 max possible tonal values
+3 2048 1024 = 2048
+2 2048 1024 512 = 2048
+1 2048 1024 512 256 = 2048
0 2048 1024 512 256 128 = 2048
-1 1024 512 256 128 = 1024
-2 512 256 128 = 512
-3 256 128 = 256
-4 128 = 128So if the best exposed "stop" is taken from each image, it seems that 5 shots at 1 stop apart gives more possibilities of achieving 2048 discrete tonal values per shot. That is, it has 5 stops at the full possibility of 2048 values, compared to 3 stops having 2048 values when shooting 3 shots 1 stop apart.


This is why I think 5@1 is better than 3@2, and why I think that 5@1 "provides more information" - there is more information to choose from, thus more chance for the final HDR to contain more information.

So, why am I wrong?
Again, I'm not arguing, just trying to understand. I would save a lot more time and disk space if 3@2 is just as good. I just can't understand how it possibly could be better than 5@1, which - in my thinking - provides more information for each stop.


By the way, all of this is why I think that the best range of photos to take for HDR is:
- Find the setting where the very first "blinkies" appear in the LCD.
- Back off 1/3-stop so that no blinkies appear.
- This is the darkest exposure in the range of photos
- Increase shutter speed by 1 stop
- Keep taking shots until the whole damn LCD is blinking, indicating that even the darkest shadows are over-exposed.
- Delete that shot, and keep the rest.
It seems to me that such a procedure would "expose to the right" for every stop, thus providing the maximum 2048 possible tonal values to each and every stop in the final HDR image.

Granted, this is my theory. I haven't really put this to practice.

And, in reality, it seems silly to over-expose the shadows. The human eye won't be able to tell the difference anyway. So that last 2 or 3 exposures is probably a complete waste of time.

But it makes sense to me that this procedure would provide the most possible information to the final HDR. I think.

[_GUI_]

So, why am I wrong?
Again, I'm not arguing, just trying to understand. I would save a lot more time and disk space if 3@2 is just as good. I just can't understand how it possibly could be better than 5@1, which - in my thinking - provides more information for each stop.

Hi Scottes, you are wrong in absolutely nothing, your entire explanation is brilliant. Few people understand the linear behaviour of the sensor as you demonstrated.

The problem is that having those 2048 levels of the highest stop, compared to the 1024 levels of the second stop, in practice means no improvement at all. That is why there is no reason to shoot 1EV apart compared to shooting 2EV apart, but there are reasons for not doing it as explained.

Do the following experiment:
- Shoot a scene on a tripod setting exposure to ETTR, i.e. max exposure right before clipping the highlights.
- Shoot again, but now exposing 1 stop faster.
- Take the 2 RAW files and develop them exactly in the same way (same ACR settings), only correcting the second by +1EC to make both images render in the same exposure.
- Now compare the highlights side by side.

Do you see any difference in quality, noise or postprocessing robustness? I am certainly sure that you won't.

By exposing 1 stop more you have twice the number of levels and less noise, that is correct, but you will never enjoy those extra levels since 1024 are already far more than needed for any postprocessing. And the improvement in noise is also negligible since noise is never a problem in the highlights.

One could argue that with the 5 shots we can have an improvement in the deep shadows compared to 3 shots (because there are fewer levels and more noise there), but this is again false since both bracketing schemes shared the most exposed shot (+2):

{-2,-1,0,+1,+2}
{-2,0,+2}

So both will provide the same amount of levels and good signal to noise ratio in the deep shadows.

Regards.

[Scottes]

Hi Scottes, you are wrong in absolutely nothing, your entire explanation is brilliant. Few people understand the linear behaviour of the sensor as you demonstrated.
Well thanks. At least I'm not crazy in my thinking - it seems so logical to me that I'd have serious problems if my logic was flawed.

Do you see any difference in quality, noise or postprocessing robustness? I am certainly sure that you won't.
Ah. now there's were YOU are wrong. I am a certified professional pixel-peeper with over 20 years experience peeping pixels. I will zoom in to 6400% and find the differences, no problem. :D

One could argue that with the 5 shots we can have an improvement in the deep shadows compared to 3 shots (because there are fewer levels and more noise there), but this is again false since both bracketing schemes shared the most exposed shot (+2):
Agreed considering 5@1 versus 3@2. But what about my thoughts about going just 1/3-stop short of overexposing the entire image? If one exposed the darkest shadows all the way to the right, you'd get the full 2048 discrete tonal values for the shadows, too.

[_GUI_]

An example of a 12 f-stops scene shot with {-2,0,+2} and processed twice, one taking only the end shots, i.e. {-2,+2}, and the second taking the three shots to find out how much the middle shot can contribute to the result.

The images have been scaled down using nearest neighbour to preserve the original per-pixel SNR, and even with that it's hard to see any difference. After the HDR processing and quick tone mapping, the differences are negligible.

http://img269.imageshack.us/img269/3139/comp2b.jpg

But let's go deep to the fusion process to rise the differences up. The 3 RAW files were fused according to the following maps:
(black is most exposed shot: +2, gray is middle shot: 0, and white is the least exposed shot: -2):

http://img24.imageshack.us/img24/6720/maps.gif

The places where the middle shot can contribute are those areas painted in gray (taken from shot 0) in the second map, that were obviously painted in white (taken from shot -2) in the first map. Only by strongly processing the area with curves we can see the first image was noisier:

http://img17.imageshack.us/img17/5772/compm.jpg

But those differences were hardly visible in the real image, with a regular tone mapping process. As I tried to explain, the contribution of middle shots is very low. In this case the 0 shot contributes in some way because the gap of 4EV was too wide. But in the case of {-2,-1,0,+1,+2} vs {-2,0,+2}, the -1 and +1 shots are definitively not worth at all.

NOTE: I didn't care at all of correctly processing the highlights or deep shadows, as they resulted exactly the same using both schemes since both the -2 (highlights) and +2 (deep shadows) shots were shared.

Hope the explanation is clear. The RAW files for download can be found here:
hdr1.cr2 (http://www.guillermoluijk.com/download/hdr1.cr2)
hdr2.cr2 (http://www.guillermoluijk.com/download/hdr2.cr2)
hdr3.cr2 (http://www.guillermoluijk.com/download/hdr3.cr2)


Funily, in areas (those painted black in the fusion map) where the most exposed shot (+2) can perfectly be used to obtain the final image, Photomatix mixes contributions from several of the input RAW files. That will end in a loss of sharpness because of the progressive blending if any misalignment between the shots remains, and more visible noise. In brief: Photomatix is sub-optimum in extracting the more information and quality from the input files.

Regards

[_GUI_]

Ah. now there's were YOU are wrong. I am a certified professional pixel-peeper with over 20 years experience peeping pixels. I will zoom in to 6400% and find the differences, no problem. :D

Great! but promise me you will provide me with the RAW files of the experiment. ;)
You have some pixel peeping in my previous post.

Agreed considering 5@1 versus 3@2. But what about my thoughts about going just 1/3-stop short of overexposing the entire image? If one exposed the darkest shadows all the way to the right, you'd get the full 2048 discrete tonal values for the shadows, too.

Sure, you can get those beloved 2048 for any area of any scene. The only requirement is to expose enough time. What I claim is that it is not necessary to do it in order to have an indistinguisable quality from a less exposed shot.
Anyway this consideration does not affect the discusion about the EV offset between the shots as long as the most exposed shot is shared in all the bracketings under comparision.

Regards

[CyberDyneSystems]

Careful guys, the Math Nerds have stepped up!

Awesome thread guys,. very interesting info!