NatureScapes.Net - The High Dynamic Range (HDR) Landscape Photography Tutorial

Published July 2006

Since the images were shot hand-held (using a Canon 5D), I went up to ISO 400 to keep the shutter speeds a bit faster – 1/320s, 1/80s and 1/20s. This contributes to more noise in the Photomatix version of the image; the CS2 version is noticeably smoother at full resolution. In most respects, both images examined at this down-sampled size look fairly equivalent. What can’t be seen is the processing time. The CS2 HDR conversion took much longer since I had to iterate over the parameters in the Local Adaptation method several times, dealing with problems (described below) and trying to get the look I wanted. I was able to get the sky fairly close in the end, but could not achieve quite the same results in the rest of the image, as I was able to get in Photomatix in just a few minutes.

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.

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

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.

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

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.

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.”

All content on this site is copyrighted material as indicated. Unauthorized use or reproduction is prohibited.