The original images: To make a HDR image you need to take three or more photos with different exposure values, because the normal exposure may lack some detail, which can only be recorded with a standard digital camera by over- or underexposing. The histogram is a good indicator: The middle image is quite OK, a good LDR, but the peaks to the very left (dark) and right (bright) tell us that some information is lost.

Alignment: I am proud to own a ridiculous mini tripod and my camera requires pushing several tiny buttons to alter shutter time. I never manage to take three or more shots without moving the camera and, thus, the frames have to be realigned before I can even think of something like HDR. I found HDRprep suitable, a perl script which usually finds the best alignment automatically. (The original images must have valid EXIF metadata such as JPEGs from digital cameras.)

A good way to tell whether two images are aligned is to open them in the GIMP as two layers of the same image and look at the difference: (If objects have (white) borders, they are not well aligned.)

Let’s go HDR: The actual HDR image creation procedure with Qtpfsgui is trivial. I used Floyd’s ubuntu edgy package with default settings. You can play around with the high dynamic range histogram and simple preview now. It looks pretty similar to the original “normal exposure” image, which proves I didn’t make a big mistake so far. There a two problems on the left hand side: Firstly, moving people appear as ghosts and, secondly, the alignment didn’t work for smaller branches of the trees. (It was probably windy.)

I decided to crop the image with pfsin | pfscut —left 1260 —top 0 | psout keeping the ducks but getting rid of people, cars and the trees.

Tone Mapping: I tried all tone mapping algorithms built into qtpfsgui randomly adjusting the settings. While trying to make the image as impressive as possible I wanted realistic colours. I liked the tone mapper labeled “Reinhard ’02” most. The result is not very different from the original but note how the histogram is compressed and equalized. Dark parts are brightened which certainly works better with my LCD.

Not quite what you expected? A natural look is not what HDRIs are famous for. Well, I made another try with Fattal’s alogrithm:

These images displaying Uppsala’s Domkyrkan are licensed CC-BY-SA.