Bending reality: an attempt at 180-deg HDR panorama
I have been experimenting with two distinct post-processing techniques, high-dynamic range tonemapping and panorama stitching, for a while and was pondering whether they could be combined and how. I thought it would be interesting to be able to produce an image with a 180-degree angle of view with tonal compression.
Both processes use software that require relatively little user intervention so I am curious about their combined ease of use and resulting image. I also want to highlight the weakest link in the process and see if it can be circumvented.
In order to start the exercise, I need a suitable series of images. I went back to the Geo W. Reed (even though I promised myself that I was done with that place…) because it is conveniently close to where I reside and shot overlapping pictures at multiple exposure levels using a wide angle lens. The image capture comprises 5 different views shot at 8 exposure levels from -4EV to +3EV for a grand total of 40 images.
Now comes the interesting part which relies heavily on the combined geniuses of multiple image processing specialists and software developers: for each exposure level, I would stitch a panorama and all individual panorama would subsequently be combined into a final HDR image. Here’s the worry: I had concerns that the stitching software would either have difficulties recognizing and aligning features in very dark pictures (for example, -4EV), and/or that the detection and rendering would slightly differ at each exposure level. Here’s the stitched picture at EV+0 before any other processing. The panorama software does not encounter any difficulty performing its intended function in the range of exposure levels I used. It produces a composite image with all edges aligned.
Actually, my second concern came true: the resulting stitched picture at each exposure level comes out with slightly different dimensions so I have to rely on the HDR software to align them properly.The HDR software does complain about the different sizes of the pictures and it does offer to crop to identical sizes.
Finally, I use a defishing software to perform minor distortion corrections and to crop to the final picture dimensions. Unfortunately, the HDR software did not produce a perfectly aligned image as some of the edges are discontinuous at the image boundaries. I still think it is feasible to obtain a HDR panorama and I need to revisit the process to find ways around its limitations. Here are some solutions I may explore:
- Change the boundaries between pictures (more or less overlap)
- Manually crop the pictures to identical size (this also means aligning manually so I am not sure how easy this would be)
- Try HDR from a single stiched image and processing it at various EV adjustments
- Try other HDR software packages
The whole process is easy and fairly quick to execute. I am somewhat disappointed that the result is less than perfect but I see enough potential to keep on experimenting with the various digital processes involved in this workflow. I might also try this technique outdoors without as many linear elements in the foreground as maybe it is more adequate for landscape rather than architectural applications. Histoire à suivre…
Equipment: Canon DSLR 5D Mk II, Canon EF 16-35mm f/2.8L