Photophys.com: The Science of Photography

In the February 28 blog I revealed my interest in panoramas and mirrors. Since that time, I have had a few failures and have learned about how to make more robust panoramas. Now I would like to comment on my experiences and to display couple of new results.

I have been using a Canon XTi DSLR camera equipped with a 10mm Sigma Fisheye lens and mounted on a Nodal Ninja 3 panoramic head; and at first I attempted to make spherical panoramas by collecting eight images (or sets of images for HDR). As previously described, four shots were obtained with an upward tilt of (+30 deg) and rotation about the vertical axis in steps of 90 deg. The remaining four shots were obtained with a downward tilt (-30 deg) and four shots similar to those at +30 deg but were shifted by 45 deg about the vertical axis. This can work, as demonstrated in the February 28 blog, but sometimes the images cannot be stitched properly because of insufficient overlap. The problem is that a small amount of misalignment or angular shift during capture can lead to failure. A 15mm fisheye lens on a FF (35mm) camera would give more overlap than my 10mm lens on the XTi because there is a crop factor of 1.6 rather than 1.5 as in Nikon APS-C cameras.  The equivalent fisheye lens for the XTi would have a focal length of about 9mm.

A much more robust procedure is to take 10 images as described above except with 72 deg rather than 90 deg rotation about the vertical axis between shots and a shift of 36 deg for the second set of five relative to the first set. With the Nodal Ninja head, a detent ring with click stops at 18 deg can be used to give four click stops between images. However, even with increased overlap, it is important to check the alignment frequently with the built-in bubble level. I expect that six shots about the vertical axis with zero deg tilt, supplemented with one straight up and one straight down, would work as well except when the overhead shot contains no distinguishing features and cannot be stitched into a panorama with standard software.

Let’s proceed with the next step. The 10 shots should be sufficient for a seamless spherical panorama, but there may be a problem with the down direction (the south pole). The panorama will always show an image of the pano head and the tripod. These can be cloned out or replaced with “Content-aware fill” in Photoshop when there is nothing of interest in that direction. When there is something of interest, such as a commemorative plaque, it will be necessary to take another photograph with the tripod out of the way so that the panorama can be patched.

For the processing steps, I first select images without HDR bracketing. I take the 10 RAW images into Lightroom 3 and correct an average shot for sharpness, noise, contrast, etc. The images are then synced with the corrected image, and all of them are exported as TIF files to a folder for panoramic stitching. Of course, this step could have been done in Photoshop Camera Raw or some other RAW conversion program. In the case of bracketed sets collected for HDR, there is a problem because I don’t know how to export the complete set (30 images) with Lightroom settings to an HDR program for batch processing. So in this case, I go immediately into Photomatix and select “Batch Bracketed Photos.” This permits me to select all 30 images from the camera for automatic RAW conversion and HDR tone mapping or fusion. The result is a set of 10 TIF images in a selected folder. I then take these images into photo processing software, in my case Lightroom. Again, one image is corrected/enhanced and the others are synced with it. For some reason, I have had better results stitching batch processed HDRs than with individually processed sets.

At this point there are 10 TIF images ready for stitching, and another software package must be selected. I have not had much luck with spherical panoramas in Photoshop CS4, and I have not tried CS5 for this task. My software of choice is Autopano Pro 2.5 (64 bit). I can simply select the set of images for processing. They are automatically detected and stitched into an equirectangular image. This means that guide point selection and lens corrections are taken care of. Typically, I do not need to do any editing in Autopano Pro, and I simply have the image rendered into a TIF file and saved in the panorama file folder. The equirectangular image should be carefully inspected for stitching errors and enhanced if necessary in Lightroom, Photoshop, or other programs.

The final step is the presentation of the image. I use Pano2VR (3.0 64 bit), but I am sure that there are other possible solutions. With Pano2VR I import the image and click the Modify button. This gives a low resolution preview of the VR image and permits the user to select the center point for viewing. The next step is to select Patch Input. Here one can zoom into any area of the panorama and extract a part for processing in Photoshop. The south pole can be replaced as desired with another photo or perhaps with a copyright notice, and when the patch is saved it can be reinserted into the panorama.

Now we get to a really interesting part. On the right hand side of the Pano2VR window one can select Quicktime, Flash, HTML5, or Transformation. I select Flash for the VR experience, but I plan to try HTML5 in the future. For remapping I select Transformation. There are a number of options. The Equirectangular image can, of course, be saved for printing with any desired panning angle. If Rectilinear is selected, the photographer can pick any direction and project out an image with a field of view (FOV) from a few degrees to 179 deg! This gives a chance to see how the scene would look with super wide angle lenses that don’t yet exist. The images with FOV of more than 120 deg are really amazing.

I love the mirror ball transformation. This indeed creates an image that looks like a real mirror ball, except the image must be flipped horizontally to get it completely right. Of course, a real mirror ball always has the camera, tripod, photographer, etc. in the center of the image. The Pano2VR mirror ball appears as it would be seen by an invisible eye.

I made a recent panorama with the camera positioned at the center of the gazebo in the Sarah P. Duke Gardens. At the time I didn’t think about the plaque underneath the tripod, but in viewing the panorama I realized it was needed. A few days later I was in the garden and I took the needed photograph with another camera, but that was good enough for patching. In the equirectangular representation of the panorama, the north and south poles are stretched from -180 to +180 deg. I found that this image could be cropped at the top and bottom to make an interesting panorama with three roughly equal bands, shadows, outside view, and vines as shown below.

It is surreal in the sense that no camera could make this image. A cylindrical panorama would be similar, but not the same. Notice that the position of the sun burst defines an angle and that to the right of the center the shadows from the rear indicate where the sun is directly behind the camera. One half of the image width is exactly 180 deg.

I also created a mirror ball representation and experimented with suitable backgrounds. First I panned until the opposite side of the image (the ends of the previous figure) was in the center. That should give the scene that would appear behind a real mirror ball. Then I experimented with various rectilinear mappings of it. The detail in these images turned out to be distracting. Finally, I resorted to the same trick I used in the previous panorama blog (Feb. 28). I increased the field of view until the rectilinear image consisted of lines radiating from the center. Projections with FOVs of 150 or 160 deg can be remapped fairly quickly, but as the FOV approaches 179 deg the remapping time increases to at least half an hour on a quad core machine with 8 GB RAM. For the final image I settled on the mirror ball in front of a blurred rectilinear image with a FOV of 160 deg as shown below.

One concern with the panoramas is that no lens can focus perfectly from zero to infinity simultaneously. The 10mm fisheye does rather well at f/11, but even so, the tripod and ground below the camera will not be in perfect focus with the same setting as that required for the horizon. In very precise work each image should be a focus stack as well as an HDR set. This would require processing with a program such Helicon Focus as well as tone mapping the HDR set. An extreme example of this kind of precision work can be seen here. The artist, Julian Kalmar, combined 1200 raw images taken with a 100mm lens.