Breaking the Gigapixel Barrier

(Posted: November 28, 2003)


December 14, 2003 update: In early December this page was featured on numerous popular websites (Slashdot, luminous-landscape, imaging-resource and others), and I was flooded with hundreds of e-mails as a result. Thanks to everyone who has written with comments, congratulations, suggestions and ideas. I am grateful for all the interest.

I have read all of the e-mails and am starting the process of responding. However, I will not be able to respond to all the e-mails, so I've added a "Frequently Asked Questions" section at the bottom of this page. If you don't get a response from me, it isn't personal! It is probably because your question is answered below, or elsewhere on my site.



Introduction. This page contains what I believe to be one of the highest resolution, most detailed stitched digital images ever created. It is the view from Bryce Point in Bryce Canyon National Park in Utah. It consists of 196 separate photographs taken with a 6 megapixel digital camera, and then stitched together into one seamless composite. The final image is 40,784 x 26,800 pixels in size, and contains about 1.09 billion pixels...a little more than one gigapixel. I have been unable to find any record of a higher resolution photographic (i.e. non-scientific) digital image that has been created without resizing a smaller, lower resolution image or using an interpolated image.

1.09 gigapixel image of Bryce Canyon
Resized version of 1.09 gigapixel image after stitching


How was it created? The first step in the creation of the image was to choose an appropriate subject. There are a number of technical issues that I had to consider that are not normally encountered when taking single images. For example, it took me 13 minutes simply to take all the photographs, and I was shooting as fast as my camera could write images to its memory card. So, I needed a subject that was relatively static. Secondly, I knew that I would have to use a very long focal length lens to take the image, otherwise the final composite would end up with an extremely wide field of view...something I didn't want. This also presented challenges due to the extremely short depth of field when using very long lenses.

The second step was to assemble the images. This was a complex and lengthy process. My normal procedure (using PTAssembler, Panorama Tools and Photoshop) was not sufficient in this case for a number of reasons because of the size and number of images I was working with. For example, the version of Photoshop that I use cannot work with images with pixel dimensions of more than 30,000. So, my solution was to modify some of the existing programs in my workflow, and write a number of new software programs to create this image.

196 component images, prior to combining into a 1.09 gigapixel image of Bryce Canyon
196 component images before stitching


Technical Details. Here are some facts and figures about this image:
  • Final image dimensions: 40,784 x 26,800 pixels
  • Number of pixels in final image: 1,093,011,200 (1.09 gigapixel)
  • Final image file format: RGB Tiff using deflate compression
  • Final image file size: 2,068,654,055 bytes
  • Number of source images: 196
  • Number of pixels in source images: 1,233,125,376 (196 images * 3072*2048)
  • Lens focal length: 280mm (equivalent to 450mm on a 35mm camera)
  • Aperture: F9. Shutter speed: 1/400
  • Number of control points in PTAssembler project: 779
  • Number of seams that were manually blended after stitching: 364
  • Horizontal field of view of final image: 63 degrees
  • Time required to capture component images: 13 minutes
  • Time required to set control points: 2 hours
  • Time required to optimize project: 2 days
  • Time required to stitch project: 4 days
  • Time required to blend seams / correct misalignments / finalize image: 3 days


How much detail does it contain? Much, much more than would be captured by any conventional digital camera...even those that cost more than a new car. For example, the Canon 1Ds (about $8,000) captures 11 megapixels, while the BetterLight Super 10K-2 scanning back (camera not included!) captures 140 megapixels, but costs about $25,000. I also believe that a gigapixel image surpasses what even die-hard admirers of large format photography argue is possible with large format cameras. For more thoughts on this subject, you might also want to read this essay.

Here's another way to think about it. Given that the resolving power of the human eye (under ideal conditions at the center of the retina) is about 1 arcminute (1/60th of one degree), this image captures considerably more detail than I (or any other normal sighted human) was able to see with my eye when standing on the overlook at Bryce Point. Assuming one pixel per arcminute, an image with dimensions of 3780 x 2485 would suffice to capture the amount of detail that the naked eye could resolve. This image has more than 100 times this detail. Looking at the full sized digital image, one is able to see things that might have been difficult or impossible to spot, even when using binoculars.

Here is full sized crop from the original image. It is 40784 pixels wide, but only 100 pixels tall. It covers the entire width of the full size image, and illustrates the amount of detail captured. It has not been sharpened, and is saved as a low quality 380KB file (to save bandwidth). You may need to save it to your hard-drive and view it using an image viewer if your browser cannot open it successfully.

Below are some crops to simulate the amount of detail that would be captured using cameras of different resolutions (I don't own any of these higher resolution cameras, so the crops below are simulated, and due to the resizing algorithm used to create these crops, they may over-estimate the amount of detail actually captured by these cameras).

Unresized crop from original 1.09 gigapixel image


140 Megapixel camera

50 Megapixel camera

22 Megapixel camera

11 Megapixel camera

6 Megapixel camera

3 Megapixel camera


Why Bother? Good question. The short answer is "why not?" As digital camera resolutions have increased, and the hardware, techniques and software for stitching multiple images into composites have improved, there has been speculation about when gigapixel images would become possible. This seemed like an interesting challenge to me. (I still think that it will be a long time before true gigapixel cameras will become available.)

However, this isn't the only reason. I've been producing and printing stitched images consisting of 20-150 megapixels for several years. I've become addicted to the amazing detail that is visible in large prints from these images! Gigapixel images present the possibility of producing some of the most amazingly detailed prints at sizes of 10-15 feet wide. A 300ppi print of this image would measure about 11 feet wide, while a 240 ppi print would be close to 15 feet wide. Even printed at this size, the image would appear very sharp upon close inspection.

Another advantage to an image this size is the ability to crop very small portions of this image in a number of different ways and still produce extremely high resolution large prints.

How Do You Print It? Another good question. The short answer to this is that there appear to be a number of alternatives, but none that I've discovered that I'm completely happy with. So, I'm still thinking about it! However, I'm interested in hearing from anyone who would like to partner with me on printing this image. I think it would be an excellent match for (and an excellent demonstration of) large format printing technology. If you have an idea or a proposal, please let me know (e-mail me).

Comments/Suggestions? Feel free to post a question or join the discussion in the forums.





Frequently Asked Questions. In an attempt to answer some of the questions that I've received since I posted this image, below are answers to some of the most frequently asked questions.
  • Where can I download the full size image? Will you e-mail it to me? I'm not making this available for a number of reasons. First, it would take you close to a week to download it over a 56K modem or about 6 hours over a good DSL connection. Second, each download would cost me (in bandwidth charges) enough money that it would probably be more cost effective for me to burn and mail you a DVD with the image. Third, giving away the full sized digital image would be similar to a film photographer giving away his negative...something I'm not quite ready to do yet!

  • The crop that I downloaded is corrupt...it won't open in my browser/image editor/favorite program. The crop isn't corrupt. Rather, many programs (including some web browers) are simply not capable of opening up images with extremely large dimensions. I'd recommend irfanview as a good freeware program that can open this crop.

  • Can't a local office/copy store print the image? My inkjet printer prints on paper up to 13 inches...I can do it. Most copy store/home inkjet printers aren't wide enough to produce a 300 or 240 ppi print of the image in one piece. They'd have to print it in a number of tiles or strips. The widest printer that I've seen promises output at sizes up to 72 inches wide...over a foot too narrow to produce a 300 ppi print of this image, and about 3 feet too narrow for a 240 ppi print.

  • Can't a billboard printer print the image? Probably. But I don't think that the image quality of these printers will do justice to this image. Remember, billboards aren't desiged to be viewed from 6 inches away. The whole point of my gigapixel image is that it contains so much detail that it can be viewed from 6 inches, 6 feet or 60 feet...it will look very sharp at any distance.

  • How did you do it? Did you move the camera? What kind of tripod do you use? etc... My general process for stitching images is outlined in my PTAssembler tutorial. However, I should warn you that I had to adapt this process significantly, and write some additional programs to work around a number of bugs in the programs that I use (interestingly, PTAssembler did not need to be modified).

    I used a Manfrotto tripod with a Bogen 3028 tripod head. I did not use a special spherical tripod head, nor did I did not use any sort of indexed or "click-stop" head to assist in precise rotation. Everything was so far away from the camera in this shot that the parallax errors introduced by not using a spherical tripod head were essentially zero. My procedure was simple...take an image, rotate the tripod head so that the image overlaps the previous image slightly, and repeat 196 times. I've been using this technique for about 5 years, which allows me to do it quickly and precisely.

  • What kind of supercomputer did you use?! Nothing too fancy. I used a generic PC with a (modest by today's standards) AMD Athlon 1800 processor and 1.5GB of RAM. The OS is Windows XP.

  • How does this compare to 8x10 or 11x14 large format? I'm not an expert in large format photography. However, in a recent post on the rec.photo.equipment.large-format (and other) newsgroup Brian Caldwell made some interesting comparisons, suggesting that this image might compare quite well to those produced by cold war spy cameras, and concluding that "ordinary large format equipment can't even begin to compare with these examples of cold war excess".

  • Can I get high resolution wall-paper of this image (or others)? I am considering selling a collection of my images as high resolution wall paper, including panoramic format images suitable for multi-monitor displays. However, I am not currently offering high resolution images for download.

  • Is the gigapixel image for sale? Once I've figured out how to print this to my satisfaction, then I may offer prints for sale. I already offer prints of some of my images (recent images only) for sale on my web site.



Home

© 2000-2003. All images on this site are copyrighted by
Max Lyons and may not
be used for any purpose without explicit written permission.