HDR is all the rage today. Wherever you turn, it's there. It's the new toy of the techno-aware photographers. And as more and more people are learning how following simple steps they can create surreal looking imagery, some crucial points are overlooked, people are failing to distinguish between the deep concept of HDR and just copying the its looks.

The myth I will try to dispel in this article: what's real HDR and what just looks like HDR. I'll also be talking about the concept of HDR, some related procedures like HDR merge and tone-mapping, when to use HDR and when not to bother, some basic prerequirements, how to shoot and bracket etc.

This article will focus on some rarely discussed issues, so what I won't be talking about is how to actually do the HDR merge once you have the images. There are literally thousands of detailed tutorials with plenty of good examples to walk you through, step-by-step. Just type "hdr tutorial" in any search engine, or look for links at the end of the article.


History

HDR is not a new invention, at least in IT terms. It was first invented as means of addressing the unrealistic lighting in CGI generated scenes, involving 3D graphics in movies, games etc.
The real world is a complex place, where a thing can be as bright or as loud as 'it wants'. To translate this world into a computer system, be it a game with 3D environment or movies special effects scene, you need to impose some restrictions. You need to choose what will be the darkest and the lightest tone in a scene, you need to make steps in between which computer can understand with it's binary language etc.
First computers used for CGI and/or 3D games had very limited resources so everything wasn't very realistic looking or took ages to render. These restrictions lingered on even as computers became more powerful -- until introduction of HDR concept, that is.
Now let's focus exclusively on HDR photography from this point on.


Dynamic range

HDR stands for High Dynamic Range. Most photographers above the level of snap-shooters should at least vaguely recognize the term, but here's a short reminder.
Dynamic range is, in very simple terms, difference between brightest and darkest tone you can perceive or reproduce (or loudest and quietest sound in audio etc.). The lower value is determined by the noise levels in a system, but that's not really that important right now.
You eye has it's dynamic range, just as digital cameras, scanners, negative and slide film, photographic paper, computer monitor, TV and even the screen on your mobile phone.
All these devices have specific dynamic ranges, your eyes being far superior to anything else. But you may recall that there are times when something is too bright or too dark even for your eyes to see, especially comparing to its surroundings. In this case that object (sun for example) is out of your eye's dynamic range, you perceive it as completely bright or completely dark (100% white or black, so to speak).

Now imagine yourself in the same situation with a camera, trying to capture that scene. Cameras have laughable dynamical ranges in comparison with your eyes, so if your eyes sometimes have problems with a contrasty scene, imagine how often your camera will have the same problems.
Chances are you don't have to imagine. Go shooting on a sunny day and it's easy to find a scene so contrasty that you have to choose whether you'll let the highlights in the scene burn out (pure white) or shadows go to total black.

Now don't get me wrong, sometimes contrasty scenes are interesting to look at, especially in artistic photography, but at other times you'd rather capture information in both, well lit and shaded areas and later decide what to do. (increasing the contrast digitally isn't as much of a problem, but decreasing it effectively is)

Just a quick note: a scene with high dynamic range is called contrasty, meaning big tonal differences between highlights and shadows. If you shot a grey wall, there would be no contrast at all and you'd get the narrowest possible dynamic range (difference between white and black on that scene would be 1:1).

During history some ways have been devised to address this issue, like films with wider dynamic range or gradient filters. For example, if you want to shoot a sunset filling top half of the picture, and some scenery filling the bottom half, you would use gradient filter, which is nothing else than a piece of glass transparent at one end and darker at the other. It would reduce contrast of the scene, making the sky darker, so you can capture the shot without getting burnt out sky or black ground.

However, none of these ways was as sophisticated and versatile as HDR. Using the right film can only extend the dynamic range a small bit and gradient filters are useful only if one distinct part (half) of the scene needs darkening. They won't be any use for most other purposes.


Basic concept of HDR

Let's say we have a scene with a bright lit face and relatively dark background. We know we can either choose to properly expose the face and get too dark background, or preserve the background and burn the face.
Example:



So here's the idea. We expose for the highlights, then for the mid-tones and finally for the shadows -- that's 3 shots. Then we take "good parts" of all images and combine them into one. (by "good parts" I mean everything that's not too bright or dark.)
This is core of the HDR concept put into simple words, and here's an example of what we got by merging above 3 images into one:


(increased saturation is due to tone-mapping; more on this in a moment)

What we've effectively done is expanded the dynamic range of our capture device. We used a method which enabled it to capture darker and lighter tones than it could earlier.
If our device could normally capture 8 stops (measuring from darkest to lightest tone), and we took a shot which was underexposed 2 stops, one normally exposed and one overexposed 2 stops, the resulting combined photograph will contain 2+8+2 = 12 stops of dynamic range, which is higher than original 8 stops. That's why we dub the result high dynamic range photograph, or HDR photo, and the original shots LDR (low dynamic range).


Producing HDR - now what?

There is software like Adobe Photoshop or Photomatix Pro which will do the actual combining of LDR images into an HDR image, but first you need to produce differently exposed shots. (They also need to be almost perfectly aligned, so you will usually shoot them from a tripod and in quick succession).

Upon HDR merge (using multiple LDR images to produce an HDR image), you will finish up with a 32-bit image. A few words on this.
Any camera shooting JPEGs will produce image with 8 bits per channel. This means it can differentiate 256 levels of red, green and blue in a scene. Digital SLR cameras shooting RAW will usually have 12 bit per channel color, discerning 4096 levels for each color. Advanced scanners use 16 bit/ch conversion, so they can make out 65536 different levels for each color.
However, HDR image will use 32 bits/ch in order to somehow preserve all information in this high dynamic range scene, which means that each channel of color will contain up to 4294967296 different levels of luminosity.

It's needless to say that such image will be useless as it is, because even though it contains more information about the scene than we care to have, there is no monitor or photographic paper that can display such wide dynamic range. For that matter, most image viewers can't handle such files either.

So we need to do something with the image in order to even be able to see it. What we do is called tonal compression.


Tonal compression

To put it simply, tonal compression is taking that awfully wide dynamic range and narrowing it down until it fits our intended purpose: display on monitor or print on paper.
In order to do that, we need to reduce contrast of the scene greatly, because otherwise we'd finish up with burnt out images we started from. There are ways and ways of doing this, but most simple one is simply remapping color numbers by dividing them with given factor. The results of this can be simply illustrated: take an image and decrease its contrast greatly. What you get would resemble HDR image processed this way: flat uncontrasty, even grayish image, uninteresting to look at.

Obviously, this won't do. It isn't why we bothered shooting HDR.

This is where imaging experts come in, calculating and applying tonal curves on these results to get more realistic tonal distribution, getting back a bit of contrast and color, using advanced algorithms which are to produce more realistic images (or at least more eye-pleasing). There are many parameters in these conversions, so good HDR software will have bunch of options and sliders to redistribute tones to user's desire.

It has been postulated in early 20th century that human eye is much more sensitive to relative contrast than absolute luminosity of perceived object. That's why good HDR tools incorporate something called contrast adaptation (using non-local operators), which is simply an algorithm which tries to preserve contrasts between adjacent parts of the image, at the expense of luminosity consistency throughout the image.
If this is hard to understand, look at HDR images that have bright halos around dark objects (almost always seen on sky around darker object). *EXAMPLE* This is result of preserving local contrast between dark object and light sky, but as you progress away from the object, luminosity is decreased due to tonal compression so it can fit with the rest of image. This will cause otherwise uniform color areas (e.g. sky) to be brighter near dark objects, and darker in homogenous areas away from such objects.

There are other methods for mapping 32bpc image into 16 or 8 bits/ch, but you don't have to be familiar with them as long as you understand the problems involved in this process so you can make best use of whatever's on your disposal.

This process -- tone-mapping -- is the reason why HDR images usually look surreal or even impossible. We aren't used to seeing soft transitions between contrasty objects, our brains aren't used to seeing perfectly exposed sunset alongside perfectly exposed landscape, preserved details in highlight areas as well as in deep shadows. There's something odd about these images, and that's what makes them attractive to variety of people. HDR becomes more than just a way of capturing as many details in tonal extremes as you can; it's another way of manipulating ordinary scenes into something more interesting.
And this is where we come to faking HDRs...


Fake HDRs

I've probably seen at least just as much fake HDR images on deviantART as real ones. I don't really have problems with this as long as they don't choose to call them HDR. However, most people don't even realize what HDR is nor what it's used for, so they can't discern real HDR from fake even when they make it themselves. This part of the article is aimed at them.

Most common example of fake-HDR is using tone-mapping on an LDR image (just any single image). This will redistribute tones on the image, shift contrasts, boost colors, introduce the weird HDR feeling many are after, but it will still be a low dynamic range image, only a manipulated one. (As a side note, the manipulation itself can in many cases be simulated by using simple curve tool in your imaging application.)

I've seen people taking a 8 bit/ch JPEG, squeezing it through an HDR application and posting it online as "HDR image". They don't realize that there is no expanded dynamic range, because no matter what you do to your image, information that weren't captured on the location just don't exist in the file -- image still has standard dynamic range of your camera. (And a poor one at that, because even just shooting RAW over JPEG would gain you more dynamic range than anything you could do on your computer later on). So using a single JPEG will produce extremely low dynamic range shot, hence there is no point in calling it HDR. What you might want to call it is simply "a tone-mapped image", or tone-remapped to be precise.

So why do people even do it? Because tone-mapping of the image (otherwise done during later stages of the HDR creation) alters the image in a specific way, attracting to some, so they want to immitate it.
Again, this is not a problem as long as you don't call it HDR, because it is not. In analogy, driving a pickup truck on a race-track doesn't make it a race-car.


Another thing sometimes encountered is using multiple exposures on scenes that just don't require them and are well within dynamic range of the camera as it is. It takes experience to distinguish such images, but the common denominator of all is that they depict relatively low-contrast scenes, which would have no extreme shadows or highlights in reality. Even though result of merge of such images won't produce expanded dynamic range, at least the author tried and put in some effort into it, so it's not nearly as bad as a simple HDR-fake from a single image.
This technique of using HDR on 'normal scenes' is actually sometimes used to reduce noise levels in shadows, but I rarely see someone doing it just to battle noise as it's just too much hassle.

On a side note, most HDR software will warn you if your images aren't differently exposed or their difference is insignificant (one-third stop e.g.), trying to tell you it's useless to do an HDR merge.


When to use HDR?

So lastly we come to this: how do you recognize a scene which would benefit from an HDR shot?
Well, probably the simplest way is taking a shot and seeing if your camera's dynamic range is enough to capture everything you wanted on the shot. Use the histogram feature for this: are there burnt out areas on the shot? (pure white) Can they be remedied by shortening exposure, or does this produce black areas with no detail on other parts of the image?
If you have too bright and too dark areas coexisting, then you have a candidate for an HDR merge.
Second condition: is the scene stationary and do you have a tripod? HDR images need perfect alignment, so if you're shooting a people, moving traffic or trees on a windy day, HDR isn't the ticket no matter how contrasty the scene is.

If above conditions are met, put your camera on the tripod and find an exposure where there will be no really dark tones, let the left part of your histogram be centered. This would correspond to deliberately overexposed shot, ignore the burning highlights. This will be your first HDR shot in the series.
Continue shortening exposures and shooting with 1 or 2 f-stop shifts until you shoot the final image where there will be no burnt out highlights. The right part of histogram (brightest tones) will be centered. This is a deliberate underexposure of the scene.

So now you have images ready for HDR merge. Once captured correctly, you can use variety of software and methods to do the actual merge and tone-mapping, and I leave that to you.

Here's a few links to get your started:
How to guide to HDR
Comparison of Photoshop and Photomatix tone-mapping
Bracketing technique with plenty of examples



P.S. Shot RAW if possible. You'll get even more out of your HDR that way.