You start the process of creating a tone-mapped LDR image by merging all your source images into a single 32-bit HDR image. We all know that. Maybe you have done this already very often, and probably you think that there is not much to think about when you do this. Right? Even more so as this part of the whole process is usually not covered very well in most tutorials you will find on the Internet. Well, after you have read this post, you may have to change your opinion on this. I will show you different ways of merging your 32-bit HDR image, using the tools provided by Photoshop CS4, Adobe Camera Raw, and Photomatix Pro 4.0. As you will see, there are huge differences in the results.
In my HDR work, I have come across a number of situations where especially the highlights of an image got lost somewhere in the HDR process. A close look at the source files revealed that enough details were present. However, after merging the source files, they were lost. Especially, in images like the one shown at the beginning of this post, this can be a real show-stopper since such images live from their light effects. I wanted to find out what is happening here and conducted some tests that actually solved the problem and provided me with a workflow that preserves as much highlight details (and also shadow details for that matter) as possible. In this article, I would like to share my findings with you.
Assumptions and Requirements
My assumptions are that you have produced a standard +-2EV series of three exposure-bracketed shots. Furthermore, I assume that you are working with raw photos. Having your source images only in JPEG format limits the dynamic range of your final image as I will show. Below, you see the source images I have used for testing the different methods.
It is clear to see in the source files that you should really create more exposures here since the dynamic range of the scene is a bit too large for the standard autobracketing series. This becomes apparent, for example, in the upper right of the scene where the light hits the right pillar. This section appears to be completely blown out. Of course, you may argue that I should have just used a tripod and made more source images, maybe going as low as -6EV to cover this range. And of course you are right! However, I was limited to shooting this scene hand-held, and the question is: Can you squeeze more dynamic range out of these shots or any other series of source images that you may have sitting on your hard disk that were created under suboptimal conditions? We will try to do this using the methods explained in the following.
I will compare the following 6 methods for merging an HDR
- Three RAWs: Feeding the 3 RAW files directly into Photomatix. This is supposedly the standard way and also the quickest and easiest way of doing it, and chances are that it is your way of doing it. However, you may be surprised to hear that it is not the way that HDRsoft (the company that created Photomatix) recommends!
- Three TIFFs: Converting the 3 RAW files into 3 16-bit TIFF files using Adobe Camera RAW (version 5.6) and feeding the resulting TIFFs into Photomatix. This is what HDRsoft recommends. The obvious reason for doing this is that you can correct Chromatic Aberration (CA) much more effectively in Adobe Camera RAW (or any other RAW converter software) than in Photomatix. When you look at the results of this test, you will ask yourself why HDRsoft recommends this!
- Five TIFFs: Converting the 3 RAW files to 5 TIFFs using Adobe Camera RAW (three original exposures plus one with -4EV and one with -6EV created from the original -2EV exposure) and feeding the resulting TIFFs into Photomatix. With this method, we artificially create two additional exposures using the exposure setting in Adobe Camera RAW. This is similar to the method being used for creating Pseudo-HDRs.
- Three JPEGs: Feeding 3 JPEGs directly into Photomatix. This is how we probably all started out before we discovered that RAWs offer more flexibility. This method is also still recommended by well-known HDR artists.
- Five JPEGs: Feeding 5 JPEGs directly into Photomatix – the 3 original images plus a -4EV and a -6EV version created from the original -2EV JPEG using Adobe Camera RAW (same process used for the Five TIFFs method).
- PS RAWs: Creating the 32-bit HDR using Photoshop’s “Merge to HDR” feature, saving the result in EXR format and feeding it to Photomatix.
After each of these procedures, the resulting 32-bit HDR was tone-mapped in Photomatix Pro 4.0 using the Detail Enhancer option and the following settings:
- Strength: 90
- Color Saturation: 46
- Luminosity: 10.0
- Microcontrast: 7.8
- Smoothing: 8.0
- White Point: 0.250%
- Black Point: 0%
- Gamma: 1.00
- Temperature: 0.0
- Saturation Highlights: 0.0
- Saturation Shadows: 0.0
- Micro-smoothing: 5.7
- Highlights Smoothness: 30
- Shadows Smoothness: 0
- Shadows Clipping: 0
The white balance was untouched in all images. The camera measured 4150 Kelvin. Wherever possible (e.g. in Photomatix’ RAW processing options), this setting was used explicitly. No other post-processing has been applied. The JPEGs were created using the highest resolution and the lowest compression possible with the camera model (Nikon D90 – JPEG Fine mode).
You may click on any of the images to open a larger version that allows for a more close-up inspection.
There are a few things that become apparent from looking at these results:
- The white balance is very different. Despite the fact that the same setting was used, Photomatix introduces a clear magenta cast when directly fed with RAW images in this case.
- Photoshop’s “Merge to HDR” function applies an extreme reduction in color saturation. No processing whatsoever was applied apart from merging the files and saving the result in the EXR format. The desaturation goes to a point where even subsequent post-processing cannot recover the colors entirely.
- The preservation of the details in the highlights is very different between the approaches. While the findings above are already striking and, to some extent, unexpected, this fact makes all the difference here. After all, preserving details in the highlights (and in the shadows) is the most important point of using HDRI technology in the first place. In the following, I will take a closer look at this.
Preservation of Details in the Highlights
Below, you see crops from all the resulting images. They show the light falling on the pillar at the right top of the scene. In this particular case, these very highlights are what gives life to the scene. Therefore, it is very important that they are rendered with as much detail as possible.
Which conclusions can we draw in terms of highlight preservation from this test?
- A straight-forward conversion of the RAW files into TIFFs (Three TIFFs method) clearly leads to a loss of highlight details. Apparently, Photomatix is able to draw more details from the original RAW files than from the corresponding three TIFFs.
- If there is any difference between using the converted TIFFs (Three TIFFs method) and using the JPEGs straight out of the camera (Three JPEGs method), then the Three JPEGs method preserves slightly more highlight details. This is a surprising result as the 16-bit TIFFs should contain more dynamic range, at least in theory. Note however, that upon a closer inspection of the entire image, one can see that the Three TIFFs method results in slightly more details while the Three JPEGs method produces noise that is harder to remove due to the JPEG compression artefacts. Hence, the conclusion should not be that JPEGs are the better alternative here.
- Creating additional exposures with -4EV and -6EV actually pays off, contrary to the general assumption that you cannot get more details than present in the RAW files. This statement is certainly true, but Photomatix is unable to really use those details, unless you explicitly provide them in additional exposures created through the RAW conversion. Among all the resulting images, the Five TIFFs method certainly results in the most pleasing appearance of the highlights. There are spots that are ultimately blown-out. However, the transitions are smooth and close to what the eye would see in the real scene (at least in this case).
- Apart from the strange desaturation effect, The PS RAWs method comes second in terms of highlight preservation in this test. It seems that the Five TIFFs method preserves slightly more details, but the difference is very small. It seems that Photoshop CS4 does a better job at preserving details in the highlights than the market leader Photomatix, which came as a surprise to me.
- Finally, the Five JPEGs method produces an image that clearly shows the difference between using RAW source images and using JPEGs. The blown-out areas have very harsh transitions, creating an unnatural and unpleasant appearance. This is a clear result of the JPEG compression. In fact, visually, the Three JPEGs method is superior to the more burdensome Five JPEGs method as it glosses over the loss of detail having larger areas of lost highlights with smoother transitions.
The fact that Photomatix cannot pull out all the details from the RAWs (bullet no. 3) is very surprising. I suspect, that the reasons are twofold:
- Converting your RAW files manually gives you much more control over their appearance and lets you decide what you want.
- More importantly, though, HDRsoft themselves say that their conversion technology is inferior to that used in specialized products like Adobe Camera Raw. It seems this gap in performance is significant. Do not get me wrong here! Photomatix is a superb product, but there are a few things HDR soft should work on. RAW conversion quality is one of them.
Final Conclusions and Recommendation
As for any such test, the results should be taken with a grain of salt. They may be different for different scenes and lighting conditions. However, the results indicate the following:
- Feeding the RAW files directly into Photomatix produces suboptimal results as Photomatix is unable to pull all the details from the source files.
- A mere conversion of the RAW files to TIFFs in a RAW converter looses even more details and, at least in terms of highlight preservation, is worse than using the RAWs directly. Note that there are other good reasons for the conversion!
- If you use JPEGs for your HDRs, you will inevitably lose details! This may not be an issue in many images, however, it becomes apparent whenever your source images do not cover the dynamic range perfectly. This is the case often if you are shooting hand-held. Therefore, you should tell your camera to produce RAWs!
- The Five TIFFs method is the clear winner of this test. Pulling out the last bit of detail with a RAW converter before you merge the images into an HDR image does pay off. The reason is most likely that tools like Adobe Camera RAW are specialized in doing this while Photomatix and other HDR tools use inferior software here (my personal hypothesis).
Based on these tests, I recommend to you to do following:
- Shoot in RAW format.
- Merge your HDRs using Photomatix and not in Photoshop to preserve the colors.
- Convert your RAW files to TIFFs using a high-quality RAW converter to retain control over highlights and chromatic aberration.
- If you have a scene with important highlight regions and your coverage of the dynamic range is close to the edge, you should develop additional TIFFs with a lower exposure value.