A Billion Infrared Pixel Image and More
Published: November 2007
First revision: February 2014
Stitching and infrared
I have been shooting landscapes for many years and the move to large format photography was in some sense to be expected. However, I have never been an Olympic athlete and as time goes by strength and fitness do not typically improve, at least for me!
The need to be able to produce large scale images but with a relatively light setup moved some of my photography toward stitching (not all of it, I still go around with my 8x10 and 14x17 cameras). The fascination for infrared photography that I have carried from my 120 film years received a major boost with the advent of permanently IR-modified DSLRs. Stitching and IR photography got married soon thereafter: I consider the combination of the two quite compelling, and I plan to keep pursuing it. Many images in the galleries in this site are indeed stitched images. I do not point it out explicitly because stitching is just a tool, and as much as I stay away from reporting focal length, exposure times etc (who cares), I also do not talk about IR pictures that have been taken with film or digital, or have been stitched together or are a single shot.
I make an exception to the above in this short contribution because the picture I want to show is a bit different for a number of reasons. First, it challenged my computing resources because it consists of 240 frames taken with a Rebel XT permanently modified for digital IR photography. Let me make one thing clear: I am not competing in any race for the largest publicized stitched image (which, as of May 2007, amounts to 13G pixel, and I am sure will soon be surpassed by something even bigger).
The reason I present this image, that tops a humble 1G pixel count, is simply to let people know what one can do with a four year old PC with 2G of memory (and 90GB of free hard disk to be used as temporary storage by the software) in 2007. Second, these very large images of landscapes present a series of problems when stitching them that are seldom acknowledged and the solutions of which are even less known. Third, the image has been taken in a less than convenient environment, to prove again the value of stitching as one of the many weapons in the photographer's arsenal.
The image has been taken at 10,000 feet altitude in the Italian Alps. On the far left side there is the Ortles, which almost tops 4000m. The panorama includes several other lower mountains still in Italy, while on the far right there is Austria. The tiny lake that I have marked with a red square is still in Italy but the border with Austria is a few chilometers away.
The image is a stitch of 240 frames and consists of 60,000 x 17,600 pixels. You figure out how big the wall in your house or apartment has to be to hang it on it. It is still in Adobe RGB to be able to experiments with false colors, so the file size is 3GB.
The image shown here is a much reduced version, and I have applied only some sharpening and fixed the levels. The large image will undergo more extensive manipulation, including selective sharpening, curves, some noise reduction, and dodging and burning to fix some shadows that are blocked and deal with a few high-key areas. Given the size of the file and the power (or lack thereof) of my PC I am prepared to be very patient...
To show the resolution of the image I have included a 100% crop of the tiny lake that I had marked in red and it is shown below.
It is possible to see some digital noise on the surface of the lake. This can be at least in part removed with one of the many noise reduction software packages available.
The major challenge in the creation of this panorama has certainly been in the stitching together of the images. In fact, the thing that I am most proud of is that carrying all the gear to 10,000 feet has not been such a big deal. My 8x10 equipment would have been a different story, though.
Anyway, back to stitching. The problem in stitching together very large panoramas is anything large-scale that moves, like surface water or the sky. Clouds have this annoying passion for changing shape while frames are being captured, and up on the mountains things get only worse. The result of this is a breakdown of the stitching software (at least of all the ones that I have tried) that creates objectionable blends between frames and literally invents very visible artifacts that indeed do not exist in the original image.
The image has been generated by my current stitching software, i.e., PTGui. This is an exceptionally good application, considered by many to be the best around right now. The stitching of this image took about 14 hours (wall-clock time) on my old PC. This did not give me a lot of opportunity to experiment with different ways to work around the problem of stitching the sky, but I did try a few without any success. Eventually I gave up and decided that a new approach was necessary: using good-old “divide and conquer” I stitched together all immovable objects first and then took care of the sky.
I used PTGui to put together the whole image consisting solely of frames that had at least some small still object in it. That resulted in an image of all the mountains chains and some portion of the sky. I then generated three strips of sky from the original frames going from the peaks of the mountains to the top of the final image. I will not go into details here but PTGui could not be used and I went back to a stitching software I used a few years ago, Panavue. This software is less powerful and sophisticated than PTGui but offers total manual control on stitching with a simpler, faster user interface than that of PTGui. When I had the three strips of sky I opened Photoshop and layered the four images together (mountain range and three strips of sky), blended them in and took away many artifacts that happened during the stitching of dark portions of the sky.
After about 60 hours of serious work and two months into the project I had my image ready to be fine tuned, which will be my next step. After that I will figure out how to print it without breaking my piggy-bank in tests after tests or in getting some banding in the last foot of paper, after having gone successfully through fifteen feet already!
Since 2007 I witnessed a predictable race towards “my picture has more Mpixels than yours.” Knowing what to look for I went immediately to check large portions of sky only to find horrible artifacts more often than not. If only they had read this paper before … just kidding!
One a more serious note, PTGui has continuously improved, and it is still a top-level reference software package for stitching. Congratulations to the team: staying on top — and for seven years — is perhaps even more difficult than reaching it.
Lastly, larger sensors and much greater computing power have made the generation of very large pictures a far simpler task than what was five or (ouch!) ten years ago.
I have included a number of large-scale IR stitches with the latitude and longitude of the point where the picture has been taken. They are all in the Italian Alps and are in false colors. When these images are seen on the screen via a tool like Zoomify the result is certainly fascinating.
(c) 2014 Marco Annaratone, all rights reserved