The High Dynamic Range (HDR) Landscape Photography Tutorial

This article was originally published in July, 2006, and is being featured again with its original content.

As a wildlife and landscape photography enthusiast with a couple of years of serious digital shooting under my belt, I do not claim to be an expert with High Dynamic Range (HDR) imaging or photography in general. But I have fun in the field, enjoy learning as much as I can about the art and science of photography, and have produced some images that are personally rewarding, as well as enjoyed by others. I currently derive particular satisfaction from working with stitched panoramas taken at sunrise or sunset, printed on roll stock.

Table of Contents

Section 1: Overview The Situation A New(-ish) Approach Section 2: What is HDR? Definition of HDR HDR vs. 8- or 16-bit Formats Capturing HDR Image Data What Is HDR Good For? Section 3: Setting Up the Input Images Physical Setup Camera Setup Determining the Exposure Sequence RAW Conversion Single Frame Scenes vs. Multi-frame Stitched Panoramas

Section 4: Processing a Single Frame HDR Image Tools Used Workflow 1 – Photoshop CS2 Workflow 2 – Photomatix Pro Comparison of Workflow Results Section 5: Processing a Multi-Frame Stitched HDR Image Tools Used Differences from the Single Frame Workflow Workflow Overview Section 6: Conclusion Wish List References

1. Overview

Late in 2005 I began adding HDR processing into my workflow. This was done to gain greater access to the tonality present in wide and dramatically lit vistas. I mostly bypassed the usual digital exposure blending route as it seemed labor intensive, although I know the technique can produce results. Naturally I posted several HDR images to Naturescapes.net (NSN), and several people expressed interest in the technique used to create these images. At the request of the NSN editorial team, I organized my learning and thinking about HDR, and this article is the result.

For at least a few of those who read this, I hope for two things. First, I would like to add some fuel to your own creative fires in working with HDR images. Second, I hope you will post your results and share questions, ideas and techniques that work for you. There is still much to learn as this new imaging capability, its tools, and our creative use matures.

The Situation

Say I have an image that looks like this:

I captured the image at sunrise, a great time to be out in the field. My senses soaking up everything before me, I tripped the shutter, hoping to capture an image that would evoke wonder and appreciation – a hint of the moment.

Back at my workstation, I eagerly began sorting through the captures. However, despite the presence of a fair amount of dramatic light and lots of interesting tonality and detail across the original scene, images like the one above just do not evoke the experience. The clouds lack drama, detail and color; portions of the sky are far over-exposed; distant trees have turned to a muddy blur; and the ice does not reveal the snappy surface detail it showed in the early morning glow.

Of course, I realized while out shooting that there was a lot of contrast (or “dynamic range”) in the scene, and that the camera could only capture a small subset of that range. So I shot different exposures (“bracketing”), some optimized for the sky, some for the foreground ice, others for the far, shadowy trees. Not surprisingly, none of these single images really grabs me upon review.

I considered that I could use a graduated neutral density filter in situations like this. At capture time, these filters are used to block some light in the brightest part of the frame, often the sky. This effectively expands the captured dynamic range by one to three stops. Of course that does not help now, with images that I have already taken. And considering the irregular line of the mountains and the dynamic range reflected across the ice and water, I am unsure if filters would be workable for this scene.

Using an exposure blending technique, I could combine several digital files with different exposures of the scene. It seems worth trying, so I put in some effort with three exposures taken across a 4-stop range using automatic exposure bracketing. The images are layered, luminance masked and blended in Photoshop CS2, together with some curves and contrast enhancements. This produced the following image:

This is a definite improvement, and with more work I could fine tune this image further. For example, some ghosting in the moving clouds could be cloned or masked out, more work with contrast and curves could increase the drama in the clouds, some selective saturation or white point adjustments could improve the whites of the ice and snow, and so on. The exposure blending technique is used to good effect by many photographers, but it can mean a lot of work. And I feel it will leave me wanting more from this image.

A New(-ish) Approach

Even with all of the above techniques, plus a lot of effort, the results may not have maximum impact. Perhaps you, like me, have wondered if there is another way. Enter High Dynamic Range (HDR). HDR imaging has been around for at least a couple of decades, but has been popularized more recently by new software tools.

Using one such tool, the Photomatix Pro HDR processing application, I tried again with the example image. On the same three bracketed exposures, I can produce this image:

This result better represents the feeling I had at the scene, represented by the title “Stormy Sunrise.” As a bonus, creating this image required no use of filters in the field, and minimal fine tuning work in Photoshop. The heavy lifting was done by “tone mapping,” a process of converting an HDR image back into an 8-bit or 16-bit image file that can be worked with conventionally (as the HDR image itself can not). In total it took less than 30 minutes of processing time after converting the RAW files – significantly less time than for the blended exposure version of the image, which honestly still needs more work.

The HDR image reveals more of the original scene’s drama than does the exposure blended version. More detail is visible throughout the sky, mountains and ice, in large part due to what are called local contrast enhancements – adjustments that emphasize tonal transitions and details within a very small space rather than strictly preserving the overall relationship of bright and dark tones across the entire image. Overall contrast and color tone is more expressive.

As for the original single frame with its middle-of-the-road, neutral exposure? While it could be tweaked, it is not remotely in the same league for expressing the impact of the original scene.

To see how you can use HDR as part of your workflow to create images with a large dynamic range, read on! This article gives a landscape photographer’s view of the theory behind HDR, describes how to capture the input images, and shows how to use two popular HDR tools: Photoshop CS2 and Photomatix Pro. It will also show how to use these tools to process both single frames and stitched panoramic images.

2. What is HDR?

Before getting into the tutorial, it would help to have some terms of reference. In brief, dynamic range (DR) is the range of luminance values from the darkest to the brightest. The original, real-world scene has a certain inherent DR which may be quite large – a ratio of 100,000:1 or more as DR is measured. Your eyes can perceive a subset of the scene’s DR (about 10,000:1), while your camera can record a smaller subset than your eyes can see – perhaps 400:1 for a DSLR. The DR of a monitor or a printed photograph is smaller yet.

High dynamic range (HDR) in photography means representing the full range of tonality present in the scene with high perceptual faithfulness. Most HDR techniques currently use software to combine several different exposures of a scene into a single file that maps the full range of luminance at every pixel. This HDR image is then processed in various ways depending on the ultimate usage. For most of us this means tone mapping the HDR image into a 16-bit or 8-bit digital file such as a JPEG or TIFF image.

If this is enough definition for you and you want to get into the part that shows how to get things done, feel free to skip ahead to the next section on shooting technique. The rest of this section provides the details of what HDR is for those who prefer to know “what” before getting into the “how.”

Key points covered in the rest of this section:

• Definition of HDR
• HDR vs. 8-bit or 16-bit file formats
• Capturing HDR images
• What is HDR good for?

Definition of HDR

Dynamic range (DR) is a fairly generic term used in a variety of disciplines. As described above, for our purposes in photography, DR is the range of luminance values from the darkest to the brightest. The DR of the real-world scene in front of you is the range of darkest to brightest portions available to your eye, film or imaging sensor. The DR of a camera is the subset of the scene’s DR that can be captured without being clipped on the highlight end, or reduced to noise or outright blocked up on the shadow end. Conversely, the DR of a monitor is the luminance range it can display from black to white.

High dynamic range (HDR) must mean a lot of DR. But how much is “a lot?” The standard unit for measuring luminance is candelas per square meter, or cd/m2. You may have seen this unit used in monitor specifications. According to the FAQ at www.hdrsoft.com (the web site for Photomatix Pro), “the luminance of starlight is around 0.001 cd/m2, that of a sunlit scene is around 100,000 cd/m2, [… and] the luminance of the sun itself is approximately 1,000,000,000 cd/m2.”

Without getting into the debates about which medium truly has precisely what DR, this chart summarizes some rule-of-thumb DR values for different stages of dealing with a scene:

STAGE	DYNAMIC RANGE	STOPS
Typical outdoor, sunlit scene	100,000:1 or more	~17 EV
Human eye	10,000:1	~14 EV
Film camera	up to ~2000:1	~11 EV
Digital camera	typically ~400:1	~8.5 EV
Good computer monitor	500:1 to 1000:1	9 – 10 EV
Typical photo print	100:1 up to 250:1	7 – 8 EV

One clear conclusion from this chart is that the experience of seeing the original scene, then capturing it, to reproducing it for others to see, is one of progressively losing DR. DR lost at capture time is gone for good, as it can never be regained after that point. If it can be captured as close as possible to what was present in the original scene, then perhaps something can be done to present the image to viewers with a better interpretation of the source scene’s tonality and detail.

Loosely speaking, then, HDR is the ability to capture and represent the full DR found in a scene with high perceptual accuracy and precision. To pin it down further, we need to look at digital file formats and how they represent luminance values.

Norman Koren’s web site has a good discussion of some of this information, specifically DR from digital capture through reproduction on screen or in print; see www.normankoren.com/digital_tonality.html. Sean McHugh’s web site also has a lot of good information about this subject; see for example www.cambridgeincolour.com/tutorials/dynamic-range.htm.

HDR vs. 8- or 16-bit Formats

An HDR image is represented using what can be considered a 32-bit per RGB channel format. The 32-bit numbers are decimal (or “floating point”) values, not integer values. The format records the luminosity of every point in the source scene, regardless of its level of brightness. There are several different HDR file formats in existence, including Radiance RGBE and Open-EXR. Each format encodes image data in a different way, with corresponding advantages and disadvantages.

Radiance RGBE and Open-EXR seem to be fairly dominant in terms of support in various applications. Both are supported by the tools discussed in this article. However the applications’ implementations of the formats are not necessarily compatible with one another.

Radiance RGBE was developed in the late 1980’s by Greg Ward as part of his Radiance imaging application, while Open-EXR was developed by Industrial Light and Magic and published as an open format around 2002. The key trade-off between the two formats appears to be that Radiance RGBE covers a much larger DR than Open-EXR, while Open-EXR offers more precision than Radiance RGBE. In truth both formats likely represent DR overkill for landscape shooting and most other forms of photography. But at least they provide the elbow room that is lacking in 8-bit and 16-bit formats.

The image formats with which we are all familiar, such as the JPEG and TIFF, supply a relatively small number of luminance values for each of the red, green and blue channels. Here is how the 8-bit, 16-bit, 32-bit Radiance RGBE and 32-bit Open-EXR formats break down:

	8-bit	16-bit	32-bit Radiance RGBE	32-bit Open-EXR
Maximum DR	255:1	65,535:1	1×1076:1	107,000,000,000:1
Luminance value range	0, 1, 2, …, 255	0, 1, 2, …, 65,535	0, 1×10-38, …, 1×1038	0, 0.0000012, …, 65,000

Note that I am glossing over a few things here since this is not primarily a technical article. For example, the Radiance RGBE and Open-EXR file formats do not actually use 32 bits per channel in the file saved to disk, for storage size reasons. Also the DR ratios are not “apples-to-apples” comparisons since the darkest and lightest luminance values vary widely. I am also not attempting to address color space or gamma encoding which affect the image data encoded within the file. For more information on this topic, see the paper “High Dynamic Range Image Encodings” by Greg Ward.

There are several pragmatic benefits of the HDR formats over the 8-bit and 16-bit formats. When the source scene DR exceeds the luminance values the 8-bit and 16-bit formats can represent on either or both ends of the range, the DR is clipped. Shadows below the low end of the range block up to black (luminance value 0), while highlights above the top end of the range blow out to white (luminance value 255 or 65,535 depending on the format).

In addition, luminance values within the representable range must “snap to” the integral values within the format’s range. There is no way to represent a luminance value of 2.3 or 2.9, so these values both may be represented as 2. Thus, in addition to the range being limited potentially on both ends of the spectrum, the values it can represent are not necessarily very precise. Image processing functions that affect luminance may introduce increasing levels of error.

Finally, the integer luminance values are not mapped to the actual light from the source scene in a linear fashion. Some parts of the scene’s tonal range are compressed into fewer values while other parts of the scene’s range get a larger block of values. Thus the non-HDR format effectively applies a tone curve that biases the DR. Compressing part of the DR can cause issues like posterization and banding when image processing is done later.

In contrast (no pun intended), the HDR formats do not impose a practical limit on the DR that can be represented. (There is a limit, just not one that a landscape or indeed almost any type of photographer is likely to hit.) This is not only because HDR uses more bits to represent a wider range of luminance values, but also because these bits represent floating point values rather than integer values. Thus HDR can represent very small and very large luminance values, such as 0.00001 and 1,000,000,000 within the same file. And because the HDR formats have decimal precision, they can represent luminance values such as 2.3, 2.9 or 2.543635. Image processing functions that affect luminance introduce fewer errors.

Finally, the various HDR formats’ luminance values correspond linearly to the amount of light present at each point. There is no curve applied to compress part of the DR into a limited span of values. Posterization and banding are unlikely to occur when working with HDR files.

All of this sounds like great theory – as long as there is a way to actually capture this image data and get it into an HDR format.

Capturing HDR Image Data

The ability to capture the source scene luminance information is a critical point. Just because a file format has a large theoretical maximum DR, does not mean that the image data contained in the file truly spans that DR. For example, a DSLR RAW file contains imaging sensor data that typically has 12 bits of total information, for maximum DR of 4095:1. (Very few digital cameras currently capture more than 12 bits of data at the sensor.)

Converting the RAW file to a 16-bit TIFF does not somehow expand the captured DR upwards. The RAW conversion process can not manufacture more luminance information than was actually contained within the RAW data. While the theoretical maximum DR of a 16-bit TIFF is 65,535:1, the real maximum DR of the image data it contains is far less. In fact it is exactly the same as the original RAW image data, 4095:1 at best. The TIFF data takes up more space but it does not contain any extra luminance information.

In fact, the DR of the data is even more limited, since the usable luminance range is only a subset of the original 12 bits of RAW image data from the sensor. A typical DSLR may have around 8.5 EV of usable DR, say about 400:1. (Read some of Phil Askey’s current DSLR camera reviews at www.dpreview.com to see how real-world DR performance works out with current digital cameras.) DR typically is lost in the shadows when image detail becomes indistinguishable from noise, and it is lost in the highlights when sensors cease to respond to brightness and just blow out to white – or to incorrect colors due to unequal clipping of the RGB channels.

So how can HDR image data be captured, assuming you are not using a camera such as the Spheron SpheroCam HDR that can natively capture HDR images of 26-stops? As stated above, the human eye can perceive a DR of about 10,000:1 in a single view. But the eye has a useful range greater than that – think of what you can see in bright daylight versus at night with your vision night-adjusted. The eye’s DR is like a sliding window of perception that can be moved across scenes from very dark to very bright, taking in a certain amount of DR as a subset of a much larger operating range.

A camera is the same in this respect. As stated, a good DSLR can capture a range of perhaps 8.5 stops of luminance in one image, or about 400:1 of DR. By altering the exposure to take images that range from very under-exposed to very over-exposed, a series of slices can be captured across the source scene’s DR. These slices will form a much larger DR when combined together in software.

So the mechanism for capturing HDR image data comes down to shooting multiple exposures of the scene. For best results the exposure series should cover the entire DR of the scene, properly exposed, from shadows to highlights. With this series of exposures and software to process them, a single HDR image file can be produced. There is some technique involved in capturing and processing the images, described in the rest of this article.

What Is HDR Good For?

Before getting into the shooting technique section, some final thoughts about the “what” of HDR – what kind of photography is HDR good for? Until recently, HDR has found most of its use in synthetic imaging applications (ray-tracing, 3D scene modeling, and computer generated imaging such as gaming) as well as video post-production work.

In still photography, HDR can be used to create images from scenes that possess a broad range of tonal values from shadows to highlights. This situation is typical to landscapes and other outdoor settings, since sunlit scenes may have a DR of 100,000:1 or more. Besides traditional landscapes, other outdoor images that might benefit from HDR include those with significant highlights such as strongly lit reflective surfaces. I have seen some great HDR images of cars for example, where the metal and glass have incredible “pop.” HDR also can be applied to good effect with indoor or outdoor architectural photography, where natural and artificial light combined with shadows can produce a wide DR over various materials and surfaces.

Naturally, just because a scene contains significant DR does not mean that it all must be reproduced. We have all seen many stunning landscapes or other scenes where the photographer selects exposures to clip highlights or block up shadows in a way that enhances impact. There is no guarantee that HDR technique by itself can produce a “better” interpretation of a given scene. For decades, photographers have made artistic decisions about what is truly important, composing and exposing for that, and letting the rest go. HDR provides another tool to use, but the artistic judgment remains as important as ever. (As Michael Reichmann said in his introductory luminous-landscape.com article on HDR, “I fully expect to see some really silly if not downright ugly [HDR] images in the months ahead.”)

Since HDR technique involves taking several exposure bracketed images and combining them into a single file, it works best with relatively static situations. If there is motion within the frame – such as wind blowing the branches of a tree, or an ice skater moving across the field of view – the software will create ghosts or blurriness. Interestingly, moving water does not necessarily pose a problem to HDR tools – they may introduce a pleasing blur to the water. Landscape compositions often benefit from the inclusion of water elements; feel free to experiment with HDR when moving water is in the frame.

Personally I use HDR for landscape and scenic shots including both single frame images and multiple frame stitched panoramas. This tutorial shows examples of both. Classic cases are sunrises and sunsets. Shooting panoramas at sunrise or sunset unavoidably introduces a challenging amount of DR in part because of the large field of view. HDR is a natural technique to use for many such scenes, and this is why I first began experimenting with the process.

Even with single frame images, I use HDR where the sky has a lot of interesting cloud formations with detail and tonality that I want to capture, without giving up detail in the middle or foreground areas that are much darker in tone. I am also starting to use HDR for winter scenes involving irregular mountainscapes of ice, snow, trees and rock. Here, I do not want to blow out the highlights in the ice and snow, while the darker tones of rocks and trees may contain a lot of detail that I want to show as well.

There are many other possible uses for HDR. The rule of thumb I would suggest is this: if you have a scene with a wide range of tonalities, and there is engaging detail across both shadows and highlights that supports the vision you wish to communicate, then HDR may be the right technique for the job. This is especially true if irregular form in the subject matter prohibits use of filters, and the DR to be dealt with would involve an excessive number of layers and adjustments using an exposure blending technique.

What examples can you think of? One place to start is thinking about those scenes containing a large tonal range that you have so far struggled to capture to your satisfaction. Even if you have successfully used exposure blending techniques in Photoshop, you may find HDR to be a valuable approach in similar situations.

3. Setting Up the Input Images

Before you can create an HDR image you must first capture and prepare the input images that will feed the process. As there will be several images to process even without getting into stitched panoramas, some points on setup are worth considering to produce the best source material you can for the software to handle later on.

The recommendations in this section may sound excessive, and certainly you can work more casually in some circumstances. My own goal often is to create images rich in detail that will be printed on large media, possibly several feet in length in the case of stitched panoramas. Quality issues not visible in online image posts or small prints become more readily apparent, so I take steps to avoid them from the beginning.

The condensed description of setting up is to first get stable support for the camera. Multiple exposure blending techniques are easier to use and produce higher quality results when software does not have to attempt to compensate for alignment errors in between frames, caused by camera motion. Then set up the camera so that the only settings changing during the image sequence are the specific exposure changes you need to capture the target DR.

Once those elements are configured, determine the number of images and exposure interval you want to shoot, and shoot the sequence. Back at the workstation, use a consistent RAW conversion to process the images for input to the HDR tool.

If you are already experienced in these areas, feel free to jump ahead to the first HDR tutorial section covering Photoshop CS2. The rest of this section provides more background on setting up and capturing the image sequence for those readers who may be newer to this type of photography.

Key points covered in this section:

• Physical setup
• Camera setup
• Determining the exposure sequence
• RAW conversion
• Single frame scenes vs. multi-frame stitched panoramas

Physical Setup

Stable physical support of the camera is relatively important. Because the HDR software needs to map the luminance values at each corresponding pixel from the series of input frames, it is important to have the images lined up with each other as closely as possible. Both HDR tools described in this tutorial have functions to align the input images if they are slightly off. However registration errors can still occur if the camera moves too much between images. This is particularly a problem if the imaging plane rotates vertically or horizontally, as this will cause perspective shifts in between bracketed frames for which the HDR software cannot compensate.

If the shutter speeds across the bracketed exposure range are all relatively fast and your hands are steady, you may be able to hand-hold the camera. It is a simple way to start trying HDR whether or not you already have a good tripod. As long as you are shooting only a single automatically bracketed sequence of images that does not require manually changing any camera settings, you should be able to at least get at least some good approximations of what HDR can do for you.

The best way to avoid alignment problems is to have the camera mounted on a stable tripod and head. Depending on how slow the shutter speed is across the range of exposures, and whether there are ambient sources of vibration such as wind, you may need to put more or less effort into stabilizing the camera support. Most of these measures are the same as for any long exposure shooting:

• Consider lowering the tripod legs. If that would block the shots, then hang a weight from the center column hook if there is one.
• Lay a bean bag over the camera and lens to damp vibration as long as this does not block any sensors on the camera body.
• Do not let the camera strap flap around in the breeze. Remove it, or wrap it securely around the tripod.
• Shoot with a remote release rather than pressing the shutter button directly, so you do not vibrate or misalign the camera.
• If you do not have a remote release, you can try using the camera’s self timer (often available with 3 or 10 second delays). The delay allows time for vibrations caused by pressing the shutter button to damp. Be on guard with this, though, because the timer also will significantly increase the elapsed time needed to capture all frames in the exposure sequence. If the light or any subject elements are changing, blurring and ghosting as well as exposure inconsistencies can result, especially for multi-frame panoramas.
• If you are manually changing exposure settings to create your bracketed sequence, or are directly pressing the shutter button on the camera, do so carefully. Even minor shifts in the camera’s orientation can create registration errors later in the HDR software.
• If shutter speeds are slow, enable the mirror lock-up function on the camera if it has one.

Camera Setup

Some basic camera settings also need to be determined. Most fundamental is the number of images and exposure interval of the sequence – whether taken by automatic bracketing or manually changing exposure between shots. Your priority is to control the camera. Regardless of how many exposures you need, put the camera in manual exposure mode and select each exposure (or each base exposure if auto-bracketing) by hand. In particular you do not want the aperture or ISO settings being changed.

Most modern cameras have an automatic exposure bracketing function. Canon non-professional DSLR’s, which I use, can take a sequence of three images bracketed up to -2 and +2 EV around a base exposure. (On the high end of the scale, the Canon 1 series cameras can be set to bracket up to seven shots at +/- 3 EV. Nikon professional DSLR’s like the D2X can bracket up to nine shots at +/- 1 EV.)

Using one of these cameras in the simplest way, you would select the auto bracketing function for +/- 2 EV (or greater if supported), select the base exposure, and then capture three images to have your set of input frames. This is a convenient and fast way to begin working with HDR.

Why immediately jump to +/- 2 EV? The goal is to capture a wide DR. The source scene easily may contain enough DR to exceed what you can record in three images at +/- 2 EV. This might amount to 12.5 EV in total for a typical DSLR, or say roughly 6000:1. Recalling that a well-lit outdoor scene may have a DR of 100,000:1, clearly 12.5 stops is not enough to capture everything from shadows to highlights. Thus if you limit yourself to a small number of exposures, you want to record as much DR in them as possible. Stepping at 1 EV likely would be insufficient. See the next section for more about choosing the exposure sequence.

A side note: The Photoshop CS2 help entry on its Merge to HDR function states, “In general, don’t use your camera’s auto-bracket feature, because the exposure changes are usually too small.” This may apply to digital point and shoot cameras, however all DSLR’s should support exposure stepping up to 2 EV or more when auto-bracketing.

If you shoot digital, I recommend that you shoot RAW for HDR work. The point is to capture DR. If you shoot JPEG out of the camera, some amount of DR has already been sacrificed within each given image as the camera compresses the sensor data (usually 12 bits) down to 8 bits. Further, the camera applies a tone curve, compressing shadow tones in order to favor highlights. Since I personally find that shadow tones contribute a lot to my HDR work, I would not like to sacrifice them. Finally, introducing JPEG compression artifacts (however slight) into the HDR process may degrade image quality.

I understand there are arguments in favor of shooting JPEG. Typically the benefits raised are to get more continuous frames, more storage capacity, and potentially minimize subsequent workflow effort in RAW conversion. However, in my opinion, these factors are not particularly relevant for HDR work, especially landscape photography as discussed here. However, if you wish to do so, you can shoot JPEG images and process them using the HDR tools described in this article.

For most other settings, the rule of thumb is to keep the camera in manual mode for each function. This ensures that nothing the camera does will vary the image sequence in a way you do not purposely intend. When shooting the sequence, you essentially want only the exposure changing to capture the range of tonality you need. Everything else should remain constant to prevent more work later, or loss of image quality.

Here are a few final points:

• For digital shooters, set the ISO to as low a setting as conditions will allow. Any noise introduced by higher ISO may be exaggerated by the HDR tone mapping process. If the source scene contains moving elements that blur at slower shutter speeds, you can try increasing ISO to get faster shutter speeds, and take your chances with noise reduction tools later.
• Ensure white balance stays the same across the images fed into the HDR process, since it is important to preserve consistent color. If you are shooting RAW, leave the camera set for automatic white balance and later convert all input images with the same white balance setting. Or you can set a specific white balance in the camera and adjust later as required.
• If you are manually bracketing any portion of the exposure sequence, use manual focus. If auto-focus is left enabled, a different focal point may be selected between one frame and the next. This will create image combination problems that cannot be resolved by the HDR software.

Determining the Exposure Sequence

How do you know what that best exposure sequence is? Clearly this depends on the source scene. It also involves how much shadow and highlight detail you decide to capture – “high” DR does not necessarily mean “all the DR there is.” Finally it depends on how many images at what exposure interval you choose to shoot with your preferred bracketing technique.

Examine the scene looking for shadow and highlight areas. In those areas, use an external light meter, the camera’s spot meter if it has one, or some test shots consulting the histogram (if you shoot digital). This will give you some potential exposure values on each end of the DR spectrum that you ultimately need to capture. You will then have to do some quick exposure math to figure out how many frames at what EV stepping you will need to capture the scene.

You can cover the scene’s DR in fewer exposures by stepping at a higher EV interval, but it is not desirable to use a larger interval than 2 EV even if your camera supports it. Stepping by 1 EV may be preferable depending on the HDR tool. As with any software function that is interpolating reality between one recorded point and another, more data points sampled closer together produce better and smoother final results.

For example, if a given scene’s highlights meter at 1/2000s at your chosen aperture and the shadows meter at 1/4s, then ten exposures taken at an interval of 1 EV would cover the range. You could possibly eliminate one or two of the exposures on the far ends of the range if you shot RAW and can depend on your camera for good shadow and highlight detail capture. Five exposures shot 2 stops apart also would do the job, at the risk of providing a bit less smooth result.

With only a little effort, you can extend the automatic bracketed approach described in the previous section to cover five shots at a 2 EV interval, even if your camera is limited to a three shot burst using auto-bracketing. Enable auto-bracketing at +/- 2 EV and select a good central exposure based on the camera’s meter. Prior to shooting, dial in -2 EV exposure compensation and shoot three frames. Now quickly dial up to +2 EV compensation and shoot three more frames. At this point you have six frames, two of which are exposed the same; discard one of them later. This leaves you with five frames covering -4 to +4 EV, roughly 16.5 stops, which approaches DR of 100,000:1. If your camera supports exposure compensation of +/- 3 EV, you can use the same technique without throwing away one of the exposures. You end up with six frames covering -5 to +5 EV, or about 18.5 stops of DR.

Here is an example showing the same scene processed twice via HDR, from two image sequences taken seconds apart. The first sequence contains three exposures covering -2 to +2 EV, while the second contains five images covering -4 to +4 EV. Both sequences have the same central exposure, and both use an interval of 2 EV. Both were tone mapped with the same parameters in Photomatix Pro, with no further retouching done.

Three Image Sequence

Five Image Sequence

In this case, there is not much difference between the resulting tone mapped images. The one based on the five image sequence is a little more contrasty due to slightly deeper blacks. For that reason it has slightly better definition in the clouds. But on the whole the two are close, and the three exposure version likely could be touched up to become even closer. For this scene three exposures at +/- 2 EV were sufficient to capture the available DR. In part this is because the sun is not directly within the frame, and there are no extreme shadows or highlights.

Outdoor scenes can have DR of 100,000:1 or more, but many have less. Learning to read the scene can reduce the need for metering and calculating a longer exposure sequence. Likewise, gaining that experience improves on the “just shoot a bunch of exposures and hope for the best” approach.

Note that by shifting the exposure range, for example by moving the base exposure up or down when using auto-bracketing, you can bias the eventual results towards the darker or lighter tones. If manually bracketing, you can choose to leave off some of the exposures on the bottom or top end of the range. You can also shoot the exposures and then simply not include them in the HDR input. This lets you adjust the mood, block up some shadows or clip some highlights for creative reasons.

Some experimentation and experience with a variety of scenes will give you a baseline for the number and stepping of exposures you need to produce pleasing results. So far I have been using three or five frames shot at a 2 EV step and have been happy enough with my results. With only three frames, some highlights and shadows are lost but often even this small a sequence produces pleasing results.

RAW Conversion

For those who shoot in RAW mode, another setup issue that must be considered involves how to configure the RAW converter. There are numerous RAW conversion applications out there, each with its strengths, weaknesses and proponents. Which one you use is not really that important for the purposes of HDR processing. What is more important is how you do the conversion.

One key factor, as previously indicated, is to ensure that all of the images in a sequence are processed with the same white balance. Tone mapping HDR images is challenging enough without throwing different color balances into the mix. Pick a representative exposure from the input sequence, get its color temperature and tint right, and then apply the same settings to the entire sequence. Do the same with any other color enhancements that you make in the RAW conversion. Many RAW converters with strong workflow support make it easy to copy settings across a series of images.

The main point of HDR tone mapping is to set levels, apply a tone curve and make contrast enhancements to the final image containing DR information compiled from all input images. Therefore it is best to avoid making any significant exposure changes to the RAW files during conversion. If the initial exposures were set up reasonably well, most of the changes that you might make to an individual input image will be trumped by the changes later made during HDR processing.

The main reason you might be tempted to adjust exposure in the RAW conversion is to shift the entire input sequence up or down, to bias the final tone mapped HDR image. As with many digital processing functions, use a light hand. This is especially true if you are increasing the exposure as this will bring out noise that the HDR processing may emphasize further. However before even trying this, recall that shifting the exposure during RAW conversion can not bring out any more real luminance information than existed in the original RAW data. Since the HDR image is going to include all of that data by mapping all of the input images, it may be less work to perform a straight forward RAW conversion, process and tone map the HDR image, and then make final exposure adjustments to the end result.

Single Frame Scenes vs. Multi-frame Stitched Panoramas

For stitching multiple frames to work well as input to HDR processing, it is even more important to ensure that camera support is stable and all non-essential camera functions are on manual as recommended above. Stitching software can compensate for a number of things that are not quite synchronized between images across the field of view. However, the more work the software has to do, the more the quality of the final product may be jeopardized. Keeping in mind that HDR tone mapping may exaggerate undesirable details in the input images, you do not want stitching artifacts to be introduced and subsequently be magnified.

One other thing to consider when planning for both stitching and HDR work on an image sequence is the shooting time factor. If you are shooting outside at sunrise or sunset for example, the light may be changing relatively quickly in your critical shooting window. The more fiddling you have to do to capture the sequence across the field of view, the more likely it is that the light quality may change perceptibly between the beginning and the end of the sequence. It is also possible that moving elements such as clouds or water will shift enough that seamless processing will be made more difficult.

Some examples of things that can help decrease the time taken to shoot the entire sequence include:

• Use automatic exposure bracketing with a larger EV interval to speed up the capture of exposures taken at each rotation point. Fewer manual exposure changes means less time taken.
• Make sure you have chosen a good focal point that works across the field of view, and disabled auto-focus on the lens. You do not want to wait for auto-focus to hunt for a new lock at each rotation point.
• Use a lens focal length that gives a wider angle, which may cut several one or more rotation points from the sequence. You can crop later for compositional reasons if you do not mind losing the resolution. The risk here is getting an increased amount of rectilinear distortion with some lenses.
• If you normally shoot vertical frames for stitching, consider whether the composition will support horizontal shots instead. This will cover the field of view even faster.
• Ensure that your exposure sequence is not filling up your camera buffer faster than it can write. Find ways to shoot fewer exposures, in order not to be waiting on the camera while the light is changing in front of you.

4. Processing a Single Frame HDR Image

Okay, enough theory! In this section, two HDR workflows will be shown that produce a single frame (i.e. non-stitched) final image from several input images. One workflow is based on Photoshop CS2, then other on Photomatix Pro. In both cases, the example will show how to initially create the HDR image file, and then how to tone map it into a final image.

Key points covered in this section:

• Tools used
• Workflow 1 – PS CS2
• Workflow 2 – PS CS2 + Photomatix
• Comparison of workflow results

The examples will produce a tone mapped image from this sequence:

0 EV (base image)	-2 EV	+2 EV

The base image was shot at 1/15s, f/22, ISO 100. The camera used was a Canon EOS 10D with Sigma 12-24mm f/4.5-5.6 lens zoomed to 13mm. The camera was tripod mounted, and the shutter was tripped with a remote release cable. The exposure sequence was taken using auto-bracketing at an interval of +/- 2 EV. The images were shot RAW and converted to 16-bit TIFF files using Rawshooter Premium at default settings except for +20 detail, +10 saturation and +3 vibrance.

Tools Used

The example screenshots in this section are taken from a desktop workstation (AMD Athlon 64 X2 4800+ dual core system with 3 GB of RAM visible) running Windows XP SP2.

Application software versions include:

• Rawshooter Premium 1.0.3 build 77
• Adobe Photoshop CS2 9.0.1
• Photomatix Pro 2.2.3

Workflow 1 – Photoshop CS2

The first workflow shows how to produce a tone mapped HDR image using only the tools found within Photoshop CS2. At the moment, CS2 probably can be considered the preeminent tool for image processing and it is “the” workhorse application for many photographers. When CS2 was introduced, it contained new support for HDR image processing and so it is likely to be the first point of contact with HDR.

There are two key functions to use. The first is Merge to HDR, accessed via the File>Automate>Merge to HDR menu in CS2, or the Tools>Photoshop>Merge to HDR menu in Bridge. This function is how you initially combine several images taken at different exposures into a single HDR 32-bit image file.

The second function is HDR Conversion, accessed via the Image>Mode>16 Bits/Channel menu in CS2 while a previously created HDR 32-bit image is open in the editor. Dropping the image mode below 32-bit triggers CS2 to get you to specify how to tone map the HDR image, in this case to get it to fit within a 16-bit DR. (HDR Conversion can also be accessed by dropping the image mode to 8-bit.)

CS2 Merge to HDR

To get the first example rolling, the first thing to do is process the above three example images into a single HDR file. With CS2 open, select the Merge to HDR function via the File>Automate>Merge to HDR menu. This brings up the following panel:

With the drop-down list showing “Files,” I have browsed to my source images and included the three of them. The drop-down can be used instead to select “Folder” which permits adding an entire folder of images, or “Open Files” which will add the files currently open in the editor.

The only other control is a check-box to enable alignment of the input images. I recommend selecting the check-box unless there is a specific reason not to. While it will add processing time, it will help ensure that any minor image alignment errors present in the input images will be eliminated, or at least minimized. If you are sure your images will match 100%, or are not worried about ghosting and blurriness from registration problems, you can uncheck the box to speed up processing.

Clicking the OK button kicks off the first stage of processing and produces a single HDR image. This process takes 15 – 20 seconds on my workstation for this example. Most of that time is spent aligning the input images. Once alignment and initial processing are done, a second large window is displayed:

This window shows a preview of the HDR image, incorporating the full DR of the original sequence. Since your computer monitor is essentially an 8-bit device, and no tone mapping has yet been done, you are able to see only a limited subset of the full DR at any one time.

The histogram on the right hand side of the window gives a snapshot of the total DR of the combined image. Each red tick mark on the horizontal axis of the histogram represents about 1 stop of DR. The white point slider beneath the histogram controls the subset of DR displayed on the monitor. Adjust the slider left and right to examine detail in different tonal areas of the image. Using the histogram, slider and image preview, it is quickly possible to tell where the interesting image detail lies and roughly how well the merge worked (including the image alignment).

Down the left hand side, thumbnails of the input image are shown with an estimate of the relative exposure of each. Each image also has a check-box controlling whether that image’s information is included in the merged image. You can toggle an individual image on and off to roughly determine the contribution its data makes to the overall look.

Be sure to leave the bit depth drop-down showing “32 Bit/Channel” in order to generate an actual HDR image. Selecting 16 or 8 bits will immediately drop the DR of the merged image by clipping shadows and highlights to fit the selected bit depth, based on the position of the white point slider. When 32-bit is selected, the position of the white point slider does not affect the saved data in any way, but the slider’s position will be remembered the next time the HDR file is opened.

The size of the preview can be adjusted using the zoom controls at the bottom if you need to examine some details more closely. When the input images have been selected and the preview examined to your satisfaction, click OK and the HDR image will be generated. This process takes only a few seconds. At the end, the generated HDR image is open in CS2, ready for further work:

Before doing anything else, you can save the HDR image in case you want to come back to the pre-tone mapped version at a later date. Several formats are available to choose from, including Radiance RGBE (which uses the extension “.hdr”) and Open-EXR (which uses “.exr”). This example saved as an .hdr file runs about 18.5 MB in size compared to about 36 MB for just one of the input images in 16-bit TIFF format. In “.exr” format, the HDR image comes in just over 13 MB in size. For a technical comparison of the HDR formats, see Greg Ward’s paper “High Dynamic Range Image Encodings” linked at the end of this article.

While many CS2 capabilities are not available with HDR files, including layers and most of the tools, some filters and a few other functions can be used with 32-bit files. Personally I am not sure what adjustments I would wish to make at this point, given the lack of layers and the inability to directly see the full image at once because of its expanded DR. Perhaps 32-bit image processing will continue to be an area of innovation in future versions of Photoshop. For now we will simply move on to tone mapping the HDR image into a 16-bit file.

CS2 HDR Conversion

In some ways everything up to this point really has just been preparation. Now it is finally time to get a concrete result – a 16-bit image that can be edited, displayed or printed just like any conventional image. However, the tone mapped HDR image will show an interpretation of the original scene in a way that is difficult (perhaps even impractical) to achieve using other means involving comparable time, effort or cost.

Following the above actions, CS2 is sitting with the merged (and saved) HDR image open in the editor. To perform the tone mapping process, you trigger the HDR Conversion function by selecting the Image>Mode>16 Bits/Channel menu item. This brings up the following panel:

CS2 permits tone mapping an HDR image using four different methods, available in the drop-down list of this panel. The first three methods are not satisfactory, at least for the type of HDR images I work with. While easy to use, they may produce useful results only in limited circumstances. This is because they provide simplistic global effects, with few or no control points, for compressing the DR of the 32-bit image into the target bit depth.

As you select each method, the HDR image open in CS2 will show a preview of what the conversion will look like. Feel free to explore these methods on your own HDR images, but they will not be discussed further in this article.

CONVERSION METHOD	DESCRIPTION
Exposure and Gamma	Two sliders for exposure and gamma values. Lets you manually adjust the brightness and contrast of the HDR image.
Highlight Compression	No controls. Compresses the highlight values in the HDR image so they fall within the luminance values range of the 16-bits-per-channel image file.
Equalize Histogram	No controls. Compresses the dynamic range of the HDR image while trying to preserve some contrast.

The interesting conversion method is called “Local Adaptation” and this is what I will focus on. When you select this method from the drop-down list and also click the small arrow icon labeled Toning Curve and Histogram, the HDR Conversion panel now looks like this:

This provides a familiar looking curves control, with some differences from the standard curves adjustment tool. The Load and Save buttons perform a familiar function. A custom set of parameters can be set up for an image, saved (using the file name extension “.hdt”), and then reloaded as a starting point in future editing sessions. As in other editing panels, pressing the Alt key (Option key on a Mac) switches the Cancel button to Reset. Clicking Reset returns the entire panel to its default state, which means the Exposure and Gamma method with its default control values.

The panel provides no ability to select just the red, green or blue channels – the curve is applied to the entire image because HDR is about working with luminance, not color. Also, there are no black, gray or white point eyedroppers, nor any automatic functions or freehand curve drawing. The curve can be edited by moving the initial black and white end points, and by adding and dragging new points.

Editing points on the curve is one of the two main attractions of using this conversion method. It permits direct control over the translation of the HDR image’s wide DR into something that will fit within 16 ( or 8 ) bits, and also present a desirable balance of tonality and contrast. Thus editing the curve here is done for the same reason as making a curve adjustment on a normal image.

The difference here is that the curve is affecting a much greater initial DR than normal (see the red tick marks of the histogram), and will redistribute contrast as well as actually compress the DR when you hit OK. To get the look you want it may be necessary to do a more than simply apply a bit of traditional “S-curve.” The greater the DR, the more work you will have to do to prevent CS2’s automatic functions from deciding for you how to compress the DR. You may have several tonal regions that you need to adjust in different ways, rather than simply focus on a simple shadows-midtones-highlights break down as with normal images.

Looking at the unfamiliar controls on this panel, first there is the Corner check-box below the curves control. Not available with the standard curves tool, this check-box controls whether the next point added to the curve creates a smooth change along the curve where it joins other existing points, or creates a sharp and angular join. This effect can be used to create a sharp transition in the tone of the image at certain levels, rather than a transition that is smoothed out over a broader luminance range. Again, because this curve potentially must compress the DR by a large margin, you may not have the luxury of making smooth tonal transitions in some cases. Having a sharp transition can protect tonalities that you have already adjusted from being skewed by further adjustments along other parts of the curve.

As with the normal curves tool, you can move the mouse cursor over the main image window, where the pointer turns into an eyedropper. Clicking and holding over a point in the image highlights, on the curve, where that image point falls. Dragging the eyedropper around on the image, the highlight point on the curve tracks the mouse movement. This lets you target areas of the image tone you may wish to fine tune on the curve.

The principles of adjusting the curve here are the same as when editing curves in a non-HDR image. Move the bottom left and top right end points closer to the tails of the histogram, to set the black and white points where image data starts to appear. Then use one or more points along the curve to adjust its slope. Likely you often will use several points, reflecting the fact that you have more than one area of tonal range with significant detail you wish to enhance. Make the slope of the curve steeper (more vertical) to increase contrast where important image detail lies. Make the slope shallower (more horizontal) to compress tone where little or less important image detail lies. Remember that the large DR present is going to get compressed into a smaller range as a result of this tone mapping exercise, so you may have to sacrifice contrast in some areas that have good detail in order to preserve detail in other areas.

The other main attraction of this panel besides editing the curve is using the Radius and Threshold controls to alter the degree of local contrast enhancement performed. As described much earlier, local contrast enhancement increases the contrast within very small regions of the image in order to enhance the appearance of detail there. Even if you spend considerable effort editing the curve, because it is a global effect across the image it is still possible to get areas of contrast that do not work well visually. By changing these numeric values, you can exercise a different form of control over the tonal quality of the conversion.

The Radius control specifies the number of pixels that the conversion function will consider to mean “local.” A small radius, such as the default value of 16 pixels, means that contrast enhancement is applied in very tight regions across the image. This will result in hard edged, sharply defined tonal shifts that occur on a small scale. The advantage is that this tends to emphasize fine detail. The downside is that it tends to look less natural and over-processed, something that is often levied as a criticism of tone mapped HDR images.

The range of the radius is from 1 – 250 pixels. If you set the radius too low, the result will look very flat across much of the field of view because the contrast enhancements have fallen below the threshold that your eye can readily pick up. If you set the radius too high, the result often will look more natural in tonality but may seem a little “plastic” and lacking in fine details that are enhanced by strong local contrast. To determine the “right” radius for a given image, move the control between extremes that are clearly “wrong” until you find a region of values that works.

The Threshold control also affects the local contrast enhancement, and sets the difference in luminance between adjacent pixels for them to be included or excluded from the current local region. Similar to the threshold value in the Unsharp Mask filter, this control relates to edge detection. This is a key part of the human visual system’s ability to perceive detail and apparent sharpness. Changing the threshold value affects how “contrasty” the resulting image will be.

The threshold can range from 0.1 to 4.0. If you set the threshold too low, once again the converted image may appear washed out and flat. A lot of the inherent detail will be softened because very slight tonal differences between pixels are enough to exclude some of them from the region being enhanced. However the results will appear smoother and more natural. Setting the threshold too high, conversely, certainly will emphasize detail and make everything stand out. But the results likely will be too stark. I often set the threshold somewhere between 0.5 and 1.0.

Here is a small selection of images showing the visual trade-offs involved with three settings for radius, and three for threshold, all using the same adjusted curve:

SETTING	IMAGE
Radius 8, Threshold 0.5
Radius 75, Threshold 0.5
Radius 250, Threshold 0.5
Threshold 0.1, Radius 50
Threshold 0.9, Radius 50
Threshold 4.0, Radius 50

Both of these contrast enhancement settings, as well as any edits you make to the curve, are very much “season to taste” adjustments. Which values work best will depend on the scene (its breadth of DR and areas of detail) and the interpretation you intend to show. There is no “right” answer, although naturally there are many “wrong” answers.

In the end I wound up with these HDR Conversion panel settings:

These settings produced a tone mapped image looking like this:

CS2 Final Touch-up

A tone mapped HDR image can not be considered final any more than images straight out of the camera or RAW converter are final. After some consideration and editing, here is my final result:

The following adjustments were made to the converted HDR image to produce the above image:

• brightness increased slightly via an adjustment layer
• black point increased via a levels layer
• midtones punched up slightly via a curves layer
• selective brightness of trees increased slightly via a masked layer
• snowy / icy parts brightened (white point dropped slightly), via a masked levels layer
• whites of selected small areas of snow and ice increased further via a masked saturation adjustment layer
• a bit more snap added via a high pass overlay layer at 20% opacity
• noise reduction and selective sharpening on the base layer

I like the final image a lot more than the one that came straight out of the HDR Conversion. However, I am still not satisfied given the 2 hours of work that went into the image. I like the sky, and the flat gray of the midtones has been minimized. But the ice in particular is still not as punchy and detailed as I would like, and a lot of the whites have been compromised.

Working things to this point has taken quite a bit of time and involved numerous trials especially with the curve and contrast enhancement settings in the HDR Conversion panel. I do not know how much better a result could be achieved with additional work, at least at my level of expertise. In my experience so far with CS2 HDR processing this is not an isolated situation. Rather, it seems that each attempt I make gets to a certain point and then no further. Depending on the image, this point may be more or less satisfying, but generally I am left wanting more.

Instead of proceeding with more efforts to tweak this image in CS2, I will switch gears and see what results can be obtained using a different tool – Photomatix Pro.

Workflow 2 – Photomatix Pro

This second example is based on using Photomatix Pro to do the HDR processing. Other aspects of the workflow are performed as usual, including in my case Rawshooter Premium for RAW conversion and Photoshop CS2 for final touch-up.

Photomatix Pro is a stand-alone application from the French company MultimediaPhoto SARL. It may be new to you, but I suspect currently the Photomatix application is the most popular HDR tool among people actively working with HDR photography. The first version came out in early 2003, and I have been using it since late 2005. A free trial version is available; it is fully functional but applies a watermark to all generated images. Note: I have no connection with the company other than being a happy customer.

One advantage of using a stand-alone application for HDR processing is that it does not require you to have Photoshop CS2. When I began working with HDR, I was still using Photoshop Elements 3. Using Photomatix Pro, which can import JPEG and TIFF images created elsewhere and output 8- or 16-bit TIFF images in turn, was a much cheaper way to get into the game than purchasing CS2. (In the end I got CS2 anyway, one of the biggest reasons being to work with 16-bit layers, a necessity for this kind of work.) In general, I am a fan of using “best of breed” tools. CS2 is an incredibly capable application, but it is not the best at everything.

Another advantage of Photomatix Pro is that it contains several exposure processing functions besides HDR tone mapping. The tone mapper used for HDR images can even be applied to a single 16-bit TIFF image. Depending on the DR present in the original capture, tone mapping a single image can potentially add some interesting “pop”. For multi-image processing, in addition to an HDR work-up, the images can be run through more traditional exposure blending functions that do not do their work in a 32-bit HDR mode. Photomatix Pro also has some batch processing functions that can automate the process of converting several groups of input images.

If you prefer to stay within CS2, the Photomatix tone mapper is available as a plug-in that can be used in place of CS2’s own HDR Conversion function. You still use CS2’s Merge to HDR function to create the HDR image, however you then use the Photomatix plug-in to generate a tone mapped 8- or 16-bit image file. Most of the other Photomatix Pro features are not available in the plug-in, including batch processing and the “highlights and shadows” exposure blending functions.

The main focus of this section is to look at how the stand-alone Photomatix Pro application can be used as an alternative HDR processing tool. As with CS2, there are two key functions to work with. The first is Generate HDR, accessed via the HDRI>Generate HDR menu item in Photomatix Pro. This is how you initially combine several input images taken at different exposures into a single HDR image. The second function is Tone Mapping, accessed via the HDRI>Tone Mapping menu item when a previously created HDR 32-bit image is open.

Photomatix Generate HDR

The initial step is to process the same three example images into a single HDR file. With Photomatix open, select the HDRI>Generate HDR menu item. This brings up the Generate HDR – Step One panel:

I have navigated to the folder containing the input images and selected them. There is nothing else to do in this panel, so I proceed by clicking OK to get to the next step:

The purpose of this Exposure Values panel is to confirm the estimates the software made about the exposure range of the input images. If you shoot the input series with a consistent exposure interval such as 1 or 2 EV, Photomatix normally can accurately guess the relative exposure values. If it does not get them right, select the EV interval from the drop-down list, or manually enter the relative EV values beside each image thumbnail. Once these values are entered, clicking OK proceeds to the final step:

In this Generate HDR – Step Two panel, the only option I normally change is to enable alignment of the input images. As discussed previously in the CS2 example, this ensures that minor registration errors do not creep into the combined image due to any slight misalignment of the camera during the capture of the input sequence.

From the three options for camera response curve, I keep the recommended default selection, “Use standard response curve.” MultimediaPhoto recommends this since they feel modern digital camera sensors are close enough to linear in their luminance response that calibrating a curve for a specific camera is of little benefit. Furthermore, the Gamma 2.2 curve (the “standard” curve referred to by this setting) applied to the sRGB and Adobe RGB 1998 color spaces is well defined and does not introduce a need for calibration. Photomatix is unlike most HDR tools in this regard. Most of the others calculate a curve by default, and many of them require an explicit curve calibration step that creates a saved profile for the camera.

If you choose, you can calculate a response curve using the setting “Attempt to calculate response curve.” Or if you know the input images came from a linear RAW conversion, you can select “Use linear response curve.” If you calculate the curve, Photomatix does not permit you to save and recall it. It simply calculates the response from each image sequence you process; the results of this are likely to be variable depending on how many images are in the sequence and the exposure interval between them. Note: CS2 does not provide any options regarding response curve. It appears to calculate a response curve from each image sequence; like Photomatix it also does not permit saving the curve.

Upon clicking OK at this point, Photomatix will compute the HDR image. On my workstation this takes about 10 seconds, around half of which is spent on image alignment. When the HDR image is created, it is opened for viewing in the main window:

As with other applications that display an HDR image on normal hardware, the monitor does not have sufficient DR to render a complete view. With the image open in Photomatix, functions are available in HDRI>Adjust View to step the exposure up and down. This provides a similar ability to view the image as does the exposure slider in CS2. In addition, Photomatix provides a small full-resolution viewing window that can be enabled or disabled. As the mouse pointer is moved over the HDR image, the viewer shows a normalized view of the image data from a small area around the pointer. You can quickly check detail and luminance across the image without having to adjust the base exposure of the entire image.

Some DR statistics are available by selecting the HDRI>HDR Histogram menu item:

This window shows a histogram, the range of 32-bit luminance values, and a mapping of luminance values onto relative exposure values. The histogram shown for this example looks pretty comparable in distribution to the one seen in CS2, although it appears to have clipped the luminance range on the bottom and top ends where no image data exists. As you can see, this example has a DR of almost 1500:1, essentially all of which is available for use during tone mapping as the next steps will show.

This is a good time to save the HDR image in case you want to come back to it later. As with CS2, there are two principal formats: Radiance RGBE and Open-EXR. The default is Radiance RGBE. Saving in that format here, the resulting file size is about 18.8 MB. Saving as Open-EXR with ZIP compression produces a file about 13.8 MB in size. This file sizes are similar to those generated from CS2.

As noted before, even though both applications support the .hdr and .exr file formats, the way the applications encode the image data is not necessarily compatible. Thus it is best to use the same application to both encode and tone map the HDR image file rather than try to cross-process HDR files between applications. (An exception to this rule is the Photomatix tone mapping plug-in, which works on HDR images created by the CS2 Merge to HDR function.)

Photomatix Tone Mapping

Here is where the heavy lifting occurs with the Photomatix workflow. After the above steps, there is an HDR image open in the main window. I select the HDRI>Tone Mapping menu item and the following window appears:

At this point, in contrast to the CS2 approach, Photomatix provides just one function. However, this one function provides a lot of creative control. While controls may appear complex at first glance, the sliders and drop-down lists, in fact, make it very fast to iterate through the choices.

A preview of the tone mapped image is shown. For screen capture size reasons it is shown here at the minimum resolution of 512 pixels wide. Selection buttons at the lower left of the window enable widths of 768 and 1024 pixels. The preview dimension is always width; it does not adjust to the longest axis depending on whether the image is portrait or landscape.

There is no choice of arbitrary resolution or zoom ratio, such as zoom percentage or actual pixels. In part this is because the preview shown here is only an approximation of what the final tone mapped image will look like. The Photomatix developers may have felt that looking at actual pixels of a simulated view is of limited value. Once the image is generated, it can be examined at higher zoom levels, and if necessary the tone mapping can be revisited. However this cycle takes time and so flexible preview sizing is one feature that I wish was present in the Photomatix tone mapper. Even the 1024 pixel width (just added in release 2.2.3) is small to work with for large files such as stitched panoramas.

Fortunately the other tone mapping controls are to the tool’s credit. The familiar histogram is present to the left of the image preview. Unlike the CS2 histogram which shows the luminance distribution of the HDR image, this histogram shows the distribution of the final tone mapped image. As you make changes, you can watch the histogram to determine how your settings affect the distribution.

Starting at the top right and moving counter-clockwise, there are two radio buttons that allow you to select 8- or 16-bit output. The default can be specified in an application preference; I have chosen a default for 16-bit as seen here.

To the left of the bit-depth buttons is a slider labeled Strength which ranges from 0 – 100%. This control adjusts the level of local contrast enhancement. Moving the slider left produces more “natural” tones, but results look flatter and do not emphasize fine details. Moving the slider right intensifies contrast, punching up detail and adding dramatic tone. However overdoing this creates an unnatural appearance.

Below 50%, I find the Strength control produces results comparable to the CS2 HDR Conversion Local Adaptation method. At higher settings, it produces enhanced contrast results that I have not been able to duplicate in CS2. Strength defaults to 80%, which I do often find to be too high. For some images, though, having that top range may give a dramatic visual that you want.

To the left of the Strength slider is the Luminosity slider. This control brightens the image if moved to the right, or darkens it if moved to the left. Consider it to be similar to the midtones slider found in the CS2 levels adjustment. The default value is 0, and the range is -10 to +10. I find that pushing the Strength slider higher often darkens the over-all tone, so I raise Luminosity a little to compensate.

Below Luminosity is the Color Saturation slider. Defaulted at 46%, this control ranges from 20% to 100%. I find the default slightly low for my taste, but rarely adjust it by more than a few points. You may know how easy it is to overdo saturation, especially if your intended final output is a print. Photographers new to digital processing sometimes dial in too much saturation to make an image “look good,” when the issue is poor midtone contrast, sub-par black and white points, etc. HDR processing is very much about dealing with over-all image tone, enhancing local contrast, and picking good black and white points. So having to fall back on big saturation boosts to punch up an image should not be required. However, the control is here if you need it.

Immediately below the histogram are two more sliders that control the white and black points. White Clip defaults to 0.25%, while Black Clip defaults to 0.00%. I often put the Black Clip slider up a notch or two. Unlike CS2, Photomatix appears to pre-process the HDR luminance range to filter it down to a smaller subset in which image detail exists. Therefore it should be unnecessary to make big adjustments to the clipping points. But if you need to fine tune the image, the controls are here.

Below the clipping controls is the Smoothing drop-down list, containing four entries: “High,” “Medium,” “Low” and “Very Low,” with the default being “Medium.” The higher the selection, the larger the regions over which the contrast enhancements are smoothed. Smoother results look more “natural.” This control is analogous to setting the Radius value in the CS2 HDR Conversion panel, with similar effect. I normally use “High” or “Medium,” as I find the other settings look too unnatural for landscape scenes.

As an aside, one key difference between CS2 and Photomatix in this adjustment is that Photomatix provides only four discrete settings instead of a range from 1 – 250 pixels. Four settings may not be quite sufficient (six or eight would give a little more control), but it is a good approach for the types of images I work on. I honestly do not agonize in CS2 whether the smoothing radius should be 80, 82 or 83. Having that level of fine grained control adds little or nothing to the creative process for me, as I am focused on qualitative results.

This same effect often causes photographers to struggle with editing points on curves in CS2. There is a lot of precise control possible, but correlating that control to the qualitative results you want can be a challenge. Having fine grained controls is not a bad thing. However, difficulty making an initial selection that “gets in the zone” can make the tool time consuming and frustrating to use. Initial information indicates that Adobe is making improvements in this area with its Lightroom software. Hopefully software vendors continue to improve – most photographers want to concentrate on making beautiful photographs, not twiddling bits.

Below Smoothing is the Microcontrast control. Microcontrast defaults to “High,” and has the additional choices of “Medium,” “Low” and “Very Low.” I normally leave the setting on “High”. Sometimes small detail needs to be less emphasized. The application help file gives examples such as noisy input images and stitched images containing stitching artifacts. In these cases the control can be set lower. MultimediaPhoto is currently working on enhancements to this control that will permit higher settings without emphasizing image noise as much.

Finally, underneath the Microcontrast control is a set of buttons for dealing with parameter sets. The Previous button will undo the current change, while Default resets everything. The Load and Save buttons permit saving and recalling settings. Saving settings creates a small file with the extension “.xmp.” A library of these files can be built up to create a series of default looks that can be customized for individual images. This is a little harder to do with saved settings in CS2 because the curve edits are much more image-specific.

Here is a final look at the tone mapping window with the parameters I have chosen for conversion:

The key changes from the default settings include:

• Strength of 70%
• Luminosity of 2
• Saturation of 55%
• Smoothing of High

When I am satisfied with the settings, I hit the OK button and Photomatix performs the real tone mapping. This takes less than 20 seconds on my machine. The final results will look somewhat different than the preview, but I find generally they are close enough to achieve the look I want without a lot of iterations back and forth.

When the conversion completes, the final image (a 16-bit TIFF in this case) is opened in the main Photomatix window.

At this point the image can be saved for finishing work back in CS2. Here is what the final generated image looks like using the above tone mapping settings:

There are some interesting things going on here, and to my eye this is a better starting point than was the result that came directly out of the CS2 HDR Conversion function. I particularly like the ice in this version. However, the image is not finished, so back to CS2 for the final touch-up work.

Photomatix Final Touch-Up

With the tone mapped HDR image saved as a 16-bit TIFF from Photomatix, it can be loaded into CS2 for the finishing touches. When I did this the first time to create the tone mapped image seen at the beginning of the article, I did minimal touch-up work in CS2. This time around, to have a more level playing field in comparing the results between CS2 and Photomatix, I touched up both versions using similar adjustments.

Here is the final rendition of the Photomatix image:

The following adjustments were made to the converted image:

• brightness increased slightly via an adjustment layer
• black point increased slightly via a levels layer
• midtones punched up slightly via a curves layer
• brightness of trees increased slightly via a masked layer
• snowy / icy parts brightened (white point dropped slightly), via a masked levels layer
• whites of selected small areas of snow and ice increased further via a masked saturation adjustment layer
• a bit more snap added via a high pass overlay layer at 20% opacity
• noise reduction and selective sharpening on the base layer

Having gone through both workflows with the same example image, we can look at a head-to-head comparison of the results.

Comparison of Workflow Results

Here are the final images from each workflow once more:

CS2 version

Photomatix version

Certainly it is possible that either image could be improved further. However, I feel those improvements would be small increments, and not favor either version of the image over the other. This would leave the Photomatix version with the advantage. Having worked through numerous HDR images over the past six months, as well as reading a lot of commentary on the Internet, in my opinion the tone mapper in Photomatix simply produces more interesting results with less work. Here are some general critiques between the two versions:

• The Photomatix version has more drama in the sky. The CS2 version has possibly more natural tonal range (less black), but seems washed out, gray and a touch plastic in comparison.
• The Photomatix version contains better detail in the trees, especially visible from the center to the right. The CS2 version is more blocked up.
• The Photomatix version shows better contrast, color tone and detail across the ice surface. Again, the CS2 version is dull, gray and generally lacking in sharp, punchy details.
• The CS2 version has a smoother tonal appearance over-all but this comes at the cost of contrast, where the Photomatix version is better.

There is one notable downside with the Photomatix version of this image that I had to deal with during processing, remnants of which will be seen below in the 100% crops. That downside is emphasis of noise and other fine grained artifacts in the source images (such as JPEG compression or Bayer mask anti-aliasing artifacts). The Photomatix algorithms which emphasize microcontrast in desirable image detail can also really bring up the less desirable details. The CS2 version of the image does not exhibit this condition.

MultimediaPhoto is aware of this situation and the support FAQ for Photomatix mentions some potential work-arounds in addition to simply post-processing the tone mapped image with your favorite noise reduction tool. A pending release of the application is slated to include a new option for the microcontrast setting that reportedly will minimize the effects of noise. This functionality was not available in time for this article.

Here are some 100% crops from the two versions of the image:

CS2 version

Photomatix version

Actual image data looks very comparable between the two versions. The Photomatix version has more drama (darker tone and a bit more contrast), while some remaining artifacts are visible. The CS2 version is flatter in tone and a bit washed out; more work might improve it.

CS2 version

Photomatix version

The CS2 version has a bit less detail, and the trees have blocked up. Some artifacts are evident around the top edges of the up-slope groups of trees, which look like registration errors. Some whites are a bit too strong, probably clumsiness on my part. Color tone in the trees is better in the Photomatix version, but here the noise-like artifacts are again visible.

CS2 version

Photomatix version

Registration errors are again visible along the ridge line in the CS2 version. This is not adjustable in the HDR Conversion function so would have to be fixed afterwards in post-processing. The Photomatix version shows minimal hints of registration errors along the ridge line as well. The clouds and trees again are better, although the trees still show some noise.

CS2 version

Photomatix version

The CS2 version again lacks detail and tone compared to the Photomatix version.

CS2 version

Photomatix version

Color tone and apparent detail were quite lacking in the lower right corner of the CS2 version. The Photomatix version is much preferable, despite the noise.

CS2 version

Photomatix version

The CS2 version has blocked up again here and lost detail in the trees and rocks. The registration errors show up again along the ridge line. Possibly some heavy handed adjustments of mine need to be backed off. The Photomatix version is more pleasing over-all, but the noise shows here in the trees as an annoying blast of speckling.

CS2 version

Photomatix version

The open ice on the far left is smooth in the CS2 version, but again lacks punchy detail. The Photomatix version has the detail but also shows the noise again. At least here the noise is less distracting; not all noise is bad, sometimes it can add desirable texture.

When examined at the level of 100% crops, it appears that both CS2 and Photomatix have their pluses and minuses. Actual image detail present is comparable between the two versions. The major differences are in the results of tone mapping. Overall my preference for the results of Photomatix is borne out by looking at the details as well as at the whole. However, I am not trying to do a sell job on Photomatix. It is a tool like any other and if it does something you need, give it a shot. If not, use something else.

For example, while CS2’s registration errors were significant in this example, I have found cases where CS2’s alignment function did a better job than Photomatix. This might be a situation where using the Photomatix plug-in would allow the best of both worlds – HDR 32-bit file generation (including alignment) using CS2, and tone mapping using Photomatix.

Note: Photomatix Pro does have a partially manual alignment mode that may help with a troublesome series of input images. Using this mode, two control points can be specified, as with various image stitching applications, to help the alignment function figure out how to register the successive images.

Here are two more images of a different scene that provide another view of the differences between CS2 and Photomatix tone mapping. Both images were processed from a series of three exposures taken at an interval of 2 EV. Nearly identical finishing work was done on both, although the effort was not as extensive as for the main example images above.

CS2 version

Photomatix version

Despite the greater effort spent on the CS2 version, the Photomatix version has more dramatic contrast throughout thanks to the darker blacks. The snow also has much better definition than in the CS2 version. I attribute this to the microcontrast component of the Photomatix tone mapping algorithm. The CS2 version of the image simply does not hold the same level of detail in the snow.

Looking at a few 100% crops of both versions will emphasize the above points, plus provide a close look at one other critical short-coming of the CS2 version that has not been discussed so far.

CS2 version

Photomatix version

The CS2 version is a little smoother, but the Photomatix version is a little more dramatic.

CS2 version

Photomatix version

Again, the CS2 version is smoother while the grain can really be seen in the Photomatix version. When I ran noise removal on both images (around half strength on the CS2 version), I excluded the snow as I wanted to preserve maximum detail there. Some moderate noise removal in the snow would give the Photomatix version a clearer advantage due to its improved contrast, at the cost of some detail.

CS2 version

Photomatix version

The CS2 version wins for reduced noise, but the Photomatix version wins for detail.

CS2 version

Photomatix version

Here the CS2 tone mapper clearly falls down. All of the thinnest branches and blades of grass showing against the sky are ghosted. I fought with this for quite some time but the only way to preserve the branches rendered the entire image flat or introduced severe halos, due to pushing hard on the threshold setting. The Photomatix version has no problem dealing properly with luminance in the same area. Again, I attribute this to microcontrast enhancements at work. For this image and others like it that I have looked at, the CS2 tone mapper appears to have a serious drawback – it simply can not hold fine-grained detail in areas of high contrast without completely skewing the rest of the image. This can be “fixed,” but again it takes more time and effort.

Other Comparative Notes on the Tools

Here are a few miscellaneous points about the tools that may be worth noting.

CS2 enforces a minimum DR span on the input images. In some cases I have heard about, this seems to erroneously block the ability to merge files that contain sufficient DR. So far I have not personally encountered this issue. Photomatix Pro, in any event, does not attempt to force any minimum amount of DR on the images and will let you combine pretty much any sequence. In fact, you can tone map a single 16-bit TIFF image if you want to see what additional contrast and detail might be pulled out of it.

A few weeks before writing this article, I upgraded my landscape camera to a Canon EOS 5D. After a while I got through my backlog of 10D images, and started working on new images taken with the 5D. It quickly became apparent that 5D images were not being properly processed by Photomatix Pro. In discussion with the folks at MultimediaPhoto (who were very responsive), it appears that two known issues are going on. First, the Photomatix tone mapping algorithms work best when the image width and height are both multiples of a high power of two, such as 256 – 1024. In the case of the 5D, the size is a multiple of only 16 or 32, and this causes problems when the Smoothing control is set to “High.” Setting this to “Medium” instead appears to avoid the issue. Also cropping, upsampling or down-sampling the input images are possible work-arounds, if a higher power of 2 is the result.

Second, a large number of very dark pixels in the under-exposed images can bias the tone mapping algorithm to produce dark areas in the final image, or even to take the final image completely to black. Because 5D images have a resolution of 12.8 MP, they may have a larger absolute number of dark pixels and trigger the “black tone” issue. The web site has an FAQ entry on this issue and there are some work-arounds. Again, this may occur in its most extreme form when Smoothing is set to “High.”

Here is a comparison that shows the difference in processing a sequence of three 5D exposures. The input images were taken at +/-2 EV, then combined and tone mapped with default settings except for the Smoothing setting change to the second example. The third example was made by down-sampling the 5D files to 3072×2048 (the same size as 10D files). No other modifications have been made.

D File, Smoothing “High”

5D File, Smoothing “Medium”

5D File Down-sampled, Smoothing “High”

I ran into this same problem in the snowy sunset example just above, which is based on 5D frames as well. I was able to side-step the tonal artifacts in that case by cropping a little from the bottom and right of the images before combining them into an HDR file.

MultimediaPhoto is looking into both of these issues, and hopefully fixes will be coming before too long.

5. Processing a Multi-Frame Stitched HDR Image

In this section I provide an overview of my current workflow for creating multi-frame stitched HDR images. There are four main tools involved: Rawshooter Premium, Panorama Factory, Photomatix Pro and Photoshop CS2. Because of the complexity of working with everything, I will not go into details of any one tool, but will just mention some highlights at each major stage of work.

Key points covered in this section:

• Tools used
• Differences from the single frame workflow
• Workflow overview

The example will produce a tone mapped image from the following sequence of 24 frames:

0 EV (base) images

-2 EV images

+2 EV images

The base images were shot at 1/20s, f/16, ISO 400. (I did not really intend to shoot at ISO 400, but we will see how it turns out.) The camera used was a Canon EOS 5D with Sigma 12-24mm f/4.5-5.6 lens zoomed to 15mm. The camera was tripod mounted, and the shutter was tripped with a remote release cable. The exposure sequence was taken using auto-bracketing at each rotation point across the field of view, using an interval of +/- 2 EV. I did not adjust the camera to rotate around the nodal point. All images were shot in RAW mode.

Tools Used

The example screenshots in this section are taken from a desktop workstation (AMD Athlon 64 X2 4800+ dual core system with 3 GB of RAM visible) running Windows XP SP2.

Application software versions include:

• Rawshooter Premium 1.0.3 build 77
• Panorama Factory 4.3
• Photomatix Pro 2.2.3
• Adobe Photoshop CS2 9.0.1

Differences from the Single Frame Workflow

The main difference in this workflow compared to the previous ones seen for processing single frame HDR images, is that stitching work also must be done. You will need to decide whether to stitch first and work on HDR second, or the opposite. I strongly favor stitching first because HDR processing involves a lot of tone curve and local contrast optimization work. As such it should only be performed across a set of fully integrated images covering the field of view. Otherwise the tone mapping will create localized enhancements within the frames that stop at the individual frame borders. This will do poorly when fed into the stitching software.

In order to stitch first and process HDR second, it is helpful if the stitching software has the capability of stitching several image sets using identical control points. This will cut down the amount of repetitive stitch work and also ensure better registration of the images fed to the HDR tool. It does presume that there is minimal or no motion of solid objects within the field of view. If any subject matter is moving, then the stitching can not be done with identical control points. It is best in this case to fine tune each stitch to ensure a seamless job, and then worry later about how to deal with blurring or ghosts introduced by the HDR tool when it combines the stitched images.

Workflow Overview

Here are the basic steps to create a multi-frame stitched HDR image:

• Convert the RAW files
• Stitch each exposure sequence
• Combine the stitched images into a single HDR image
• Do the finishing work

I will briefly describe each step.

Convert the RAW Files

I mostly use Rawshooter Premium (RSP) for RAW conversion work. Once the administrative work is complete (organizing and naming files, and doing other asset management tasks), I call up the selected images in RSP and start working on the RAW conversion. For these example images, it looks like this:

There is less work to do in this step than usual. This is because most image exposure work is going to be done later in the HDR processing. Therefore at this point I restrict adjustments to setting a good white balance, possibly adjusting the detail and noise parameters slightly, and optionally adjusting saturation slightly.

For these example images I chose a Color Temperature of 7300 and Tint of 3, slightly warmer in color tone than the “as shot” settings straight from the camera. I also gave Detail Extraction, Noise Suppression and Saturation a boost of 10 points each to liven up the images a little. These sorts of adjustments are discretionary and I could leave them out here, choosing instead to perform equivalent actions later in the workflow.

I make these changes while zoomed in on a representative frame out of the base exposure sequence. Once I am satisfied with how that image looks, I use the RSP Copy Corrections function to transfer the same parameters to all 24 images in the sequence. This ensures consistent exposure and color tone for the images when they are later stitched and tone mapped.

It takes only a few minutes to process the RAW files into 24 TIFF images, each saved at 16 bits to maximize DR retention. Then I am ready to start the stitching work.

Stitch Each Exposure Sequence

With 24 TIFF images representing three separate exposure sequences of eight frames apiece, I will need to perform three separate stitching operations – one per exposure sequence. For convenience, I separate each group of eight images into a different temporary folder.

Using Panorama Factory (PF), I will start by stitching the base exposure sequence. If I need to manually fine tune any of the automatic results using PF’s tiling system, it is easiest to see most of what is going on in these exposures. The others are too dark or too light in key areas. That same fact may cause trouble for the automatic stitching algorithms as well, another reason to start with the base exposure sequence. Working through the wizard interface is straight forward and brings me to this point after about three minutes of processing:

Key settings I made while working through the wizard included:

• stitch in fully automatic mode
• specified lens focal length, enabled corrections for lens barrel distortion and brightness falloff
• automatically fine tune, do not sharpen
• enable exposure matching but not exposure correction (since exposure work will be done in HDR, I do not want PF doing anything to exposure other than ensuring a smooth blend)
• process the output for maximum image size

When I examine the results of the auto-stitch, I do find a lot of alignment problems, mostly within the foreground rocks. (Using a lens with little distortion helps minimize alignment problems where the software could not fully correct for lens distortion.) I correct these using PF’s tiling system which allows selective overrides of the automatic stitching. This type of work takes a long time to do, and is a big reason why it is desirable to process all stitches with the same set of control points. (In fact, if you could see the image at full scale you would realize that I did not complete the fine tuning. I had time to fix only three out of the seven overlap regions, but it is not noticeable at small resolutions.)

Unfortunately, PF does not support reusing a set of control points on another stitch sequence; this is at the top of my personal wish list for this tool. There is a “hack” that will trick the tool into processing a different set of images using a project file set up for a previous image sequence. However this trick is unsupported and may cease to work after some future release of the application. In the meantime, I use it to process the remaining two exposure sequences using the same control points established for the first sequence.

I now have three stitched images (saved as 16-bit TIFF files) as my exposures to feed into HDR processing:

0 EV base image

-2 EV image

+2 EV image

The base image actually looks decent, and with finishing work normally I might be happy to stop there. But this is an article on HDR, so the next step is to combine the three exposures.

Combine the Stitched Images into a Single HDR Image

I will use Photomatix Pro for the HDR work in this example. The process is identical to the one described for working with a single frame image; at this point Photomatix does not know or care that the three exposures being given to it are the result of a stitching exercise. Of course, the image size being much larger, everything will take longer to run. I would recommend not working with this size of image unless you have at least 2GB of RAM in your workstation; more would be better.

I combine the three images into an HDR file, and save it. (Trivia point: the 16-bit TIFF images are each 171MB in size, while the HDR image – saved in Open-EXR format with ZIP compression – is only 65MB.) I then bring up the Photomatix tone mapper and adjust the settings:

When I have something that looks good, I hit OK and generate the tone-mapped image. Unfortunately my initial intent to use Smoothing of “High” does not work on this image because of the previously described issues, so I regenerate with “Medium” instead. (This still has some tonal artifacts, for example along the lower left of the image.) Here is the result, which takes about a minute and a half to generate:

I already like it better than the base exposure image, and the finishing work has not yet been done. That comes next.

Do the Finishing Work

I saved the tone-mapped image from Photomatix as a 16-bit TIFF, and I now load that into CS2 to apply the finishing touches.

Working with this image in CS2 takes a bit of time even on my fairly powerful workstation, with the final file including layers reaching about 970MB in size. However, this highlights a side benefit of doing the majority of the contrast enhancement work in HDR (whether in CS2 or Photomatix) – working with numerous layers using an exposure blending technique would be fairly painful, given an image of this size.

For the sake of illustration, I make only a few adjustments:

• curves and levels layers to create a little more “pop”
• slight noise reduction to compensate for the ISO 400 source images (although the noise level from the 5D truly was not bad even after tone mapping)
• a combination of edge unsharp mask filter and a high pass sharpening layer

After these adjustments are done, the final image looks like this:

It took about four hours to reach this point from the start of the workflow. The single most time consuming stage was the fine tuning work in the stitcher, which took at least three hours – and would take as much again to complete.

Integrating HDR into the process added very little extra time, but produced a pleasing final image superior to anything I could produce with much more effort using blended layer techniques in CS2. The color and tonal transitions are “natural” looking to my eye, the drama in the clouds is clearly evident, and the detail is rich through-out the scene. All of this merely from the extra shooting step of auto-bracketing three exposures at +/- 2 EV, and introducing one extra workflow step of about half an hour!

If you work with stitched panoramic images covering a broad field of view, no doubt you routinely encounter much higher DR than single exposures can accommodate. While many excellent images can be created with single exposure stitches, hopefully this quick overview will encourage you to see what you can produce by adding HDR to your panorama workflow. I for one would be eager to see your results.

6. Conclusion

And now for some soap-box thoughts. It isn’t that bold a prediction to say that, one day soon, HDR technology will extend the definition of what is “normal” in photography. In the past, photographic technology has always imposed drastic limits on the dynamic range that could be captured, developed, displayed and printed. Improvements in the capability to deal with DR were marginal, and took long periods of time to occur. More success was had by photographers working within the technical limits of their equipment and media, using creativity and technique to realize compelling images despite the limitations.

Those who paint and work in other visual art forms have had centuries to grapple with and develop styles, techniques and materials that permit them to communicate their visions of the natural world using media that can not represent the full range of light in nature. In photography, HDR imaging now provides a unique capability to take a significant step towards the true DR present in natural scenes. A quote I recall seeing, but have been unable to track down, stated something along the lines of “the apprentice sees the subject, the journeyman sees the composition, and the master sees the light.” Now photographers can not only see the light but can work with it in a new way to develop their images of the natural world.

Setting aside what the photographers’ original intentions may have been, many HDR images floating around the Internet can be described as “unnatural,” “cartoonish,” “unreal” and so on. If photography is expected to portray the natural world in a way that is somehow more faithful than other visual arts, it behooves landscape and nature photographers working with HDR to find a combination of tools, technique, style and vision that does not violate the expectations of viewers. No doubt the proper place and use of HDR in nature photography will be disputed to some extent, as was (or is) the case for other techniques such as digital exposure blending, and shooting multiply exposed frames in film. There is room in the community for a variety of styles and interpretations; it keeps us all honest.

I personally feel it is legitimate to use the capabilities of HDR to present photographic images that challenge preconceptions of what a nature photograph (such as a panoramic landscape) should look like. The images I have worked on strike a chord for me, showing aspects of scenes that are “true” to my perception and evoke a “true” emotion, even if the technical facts of transformations like local contrast enhancement violate certain “rules.” HDR photographs may not qualify as documentary work, but for me they certainly can qualify as “natural world” art in photographic form.

One thing that can be counted on is that HDR tools in general are not standing still, nor is our understanding or use of them. Among the major vendors, for example, Adobe has invested in bringing 32-bit functionality into Photoshop starting with CS2 (as well as other applications in its video line). What we see in CS2 today is only the initial round of a new technology base. It is sure to improve over time. Meanwhile an advantage of smaller, more specialized companies like MultimediaPhoto and others is that they potentially can react faster to opportunities for technical innovation and requests from their customers.

Future developments in HDR tools should create more exciting opportunities for photographers. But if you are interested in working with large dynamic range in your images, I hope this article helps convince you not to wait – start today!

Wish List

I could wish for a number of things down the road, integrated within a single tool. A few items from my wish list include:

• An input image alignment function patterned on the adjustable tile system in Panorama Factory. To me, fixing alignment problems before they are “set” is preferable to fixing them at the end using cloning and healing techniques in Photoshop. More control over HDR alignment would be helpful.
• Arbitrary preview zooming like that in CS2, showing the “real” tone-mapped pixels rather than a simulation. Or perhaps a pop-up magnified viewer window (similar to the HDR image viewer in Photomatix or the small magnified view shown in most CS2 adjustment panels) showing 100% pixels after applying all tone mapping. This is somewhat limited by computing power, but I prefer to work directly on the “real” image rather than shuttle between simulations and generated actual images, iterating on changes because I can’t really see what I’m doing.
• Tone curve editing with slider type controls as in Photomatix supplemented by curve editing like CS2, showing “before and after” histograms. (CS2 shows the “before” histogram, while Photomatix shows “after.”) The “before” histogram tells me where the HDR image data is, and the “after” one shows how my tone mapping settings affect the final data distribution; both are useful. Interestingly, the initial Adobe Lightroom beta release did not contain point curve editing. Instead it introduced a slider control system for tone curve adjustment, offering quick, qualitative adjustments across highlights, shadows and midtones. Sometimes editing points on a curve, for all its precision and flexibility, does not quickly or easily produce desired results. Point editing is expected to be added into Lightroom’s curve adjustment. This would provide the best of both worlds – quick, qualitative controls to “get in the zone,” and precise controls for fine tuning. This capability with HDR tone curves would be a plus.
• Adjustable detail and noise controls in the HDR conversion function, such as what can be done with a number of RAW converters, and is present (though somewhat rudimentary) in Photomatix. The idea is to apply quality enhancements while the maximum image data is available. The HDR file is like a RAW file on steroids in some respects, and some adjustments that can be made to a normal RAW file prior to conversion make sense for an HDR image as well. Of course in CS2 a number of filters and functions can be applied to the HDR file; this may be an argument in favor of using CS2 to generate the HDR and pre-process it, then use the Photomatix plug-in to tone map it. I have not tried this yet but plan to do so.
• The ability to mask HDR tone curve adjustments to apply them to selected parts of the image. Also, the ability to apply certain level of local contrast enhancement to different degrees across different parts of the curve. Adobe may end up going in this direction if they add 32-bit layers to a future release of Photoshop. This would open up much of the selective adjustment ability that currently is not possible with either the CS2 or Photomatix tone mapper.
• The ability to combine several RAW files directly into an HDR file without having to convert to 16-bit TIFF images first. Very few adjustments need to be done when making RAW conversions targeted at HDR processing, so why not skip that step altogether? Some of the more niche HDR tools appear to support this, at least for certain RAW file formats.
• And if we are thinking of all of this, then how about an affordable DSLR that can produce an HDR image straight out of the camera? And a monitor and other output devices capable of presenting those images. These devices are being worked on, of course, but are still a ways out from being accessible to folks like me – I am sorry to say I do not have the cash for a Spheron HDR camera or Brightside HDR display.

References

Here are a few references to additional HDR tools and information that I found interesting, and which may provide you with further ideas.

Tools

• Photoshop CS2: http://www.adobe.com/products/photoshop/. The dominant image editing application, by Adobe.
• Photomatix Pro: http://www.hdrsoft.com/. A tool by MultimediaPhoto SARL for working with HDR images. Available as a stand-alone product or CS2-compatible plug-in.
• Autopano Pro: http://www.autopano.net/. A commercial tool that grew out of an original front-end for the free PanoTools, and claims to support both stitching panoramas and HDR processing in one application. I hope to test this soon.
• Radiance: http://www.radiance-online.org/. The first HDR tool, originally developed by Greg Ward. Used for ray-tracing imaging and modeling.
• HDRShop: http://www.hdrshop.com/. One of the first HDR tools, currently licensed by the University of Southern California. Originally co-authored by Paul Debevec.
• Photogenics HDR: http://www.idruna.com/photogenicshdr.html. A commercial product by Idruna for HDR drawing and image processing.
• PhotoImpact: http://www.ulead.co.uk/pi/. An image editing application by Ulead Systems which has supported HDR for at least the past two major releases. Contains some useful looking image combination features, including the ability to brush over areas from input images that will be explicitly excluded or included in the final HDR image.

HDR Tutorials

“Merge to HDR in Photoshop CS2 – A First Look:” http://www.luminous-landscape.com/tutorials/hdr.shtml. The first article and tutorial I ever read on HDR, by Michael Reichmann. It got me interested in the technique because of the way Reichmann described the potential for HDR, not just stating what buttons to press in which tools.

“HDR: High Dynamic Range Photography:” http://www.cambridgeincolour.com/tutorials/high-dynamic-range.htm. A short but useful overview of HDR in CS2, by Sean McHugh. This is the best single HDR tutorial I found in my early digging into the subject. Have a look at McHugh’s gallery. I don’t know how many (if any) of the images may have benefited from HDR work, but they are some gorgeous examples of low-light architectural photography.

“Stitched HDRI:” http://www.gregdowning.com/HDRI/stitched/. A brief tutorial by Greg Downing (not the NatureScapes.net co-founder – another Greg Downing!) He is a practitioner of HDR imaging and gigapixel panoramas, among other things. In the tutorial, he gives an overview of creating multi-row stitches of several exposure sequences, and then processing them into a tone-mapped HDR image. Tools used in his example include Realviz Stitcher and HDRShop.

“Photoshop HDR 32-bit Format: The Dawn of a New Era?” http://www.earthboundlight.com/phototips/photoshop-cs2-hdr-32bit.html. Another introductory CS2 HDR tutorial, this one by Bob Johnson.

“Photomatix Makes HDR and Blending Easy:” http://www.atncentral.com/Pages/Photomatix.htm. A very brief overview of using Photomatix for HDR image work. By Jim Lewis of Action Central.

DR and HDR Information

“FAQ – HDR image for Photography:” http://www.hdrsoft.com/resources/dri.html. Some good information from the makers of Photomatix regarding DR and HDR in digital imaging.

“Making fine prints in your digital darkroom: Tonal quality and dynamic range in digital cameras:” http://www.normankoren.com/digital_tonality.html. Some material from Norman Koren, comprehensive as usual, about DR in digital imaging.

“High Dynamic Range Image Encodings:” http://www.anyhere.com/gward/hdrenc/hdr_encodings.html. Greg Ward’s paper on HDR image encodings. A good technical read if you want a nuts-and-bolts understanding of how HDR image formats work and why, with side trips in color spaces, gamma encodings and so on.

“The Future of Digital Imaging – High Dynamic Range Photography:” http://www.cybergrain.com/tech/hdr/. An interesting, mostly non-technical article by Jon Meyer covering the motivation for HDR and a survey of some of the technology. Also provides some context from the world of painting.

“High Dynamic Range Imaging: Acquisition, Display, and Image-Based Lighting:” Morgan Kaufman Publishers, 2005. The only book currently available on HDR, as far as I know. Written by Reinhard, Ward, Pattanaik, and Debevec, several of the early researchers and developers of HDR imaging. Very technical, targeted mainly at HDR implementers rather than photographers or other users of the tools.