Breaking the Gigapixel Barrier

megas writes "Max Lyons has just posted on his site what seems to be the first 1 Gigapixel picture, created from 196 separate photographs taken with a 6 megapixel digital camera, and then stitched together into one seamless composite. According to Max, he has '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.'"

Comments

Relatively static?

[SpaceRook]

The guy said he needed a subject that was relatively static. But shadows on a canyon wall are not static. He says it took him 13 minutes. I wonder if there was any noticeable movement in the shadows in that time?

another large image

[Harald Paulsen]

It's worth to mention the WTC ground zero photo. It doesn't look to be stitched together, and it's a whopping 9372x9372, or 87 megapixels.

(using freecache to not toast my own webserver)

Re:This is not the first gigapixel image

[spiritu]

He specifically mentions scientific purposes - I'm sure he had orthophotos, other aerial photography, and satellite images in mind when he mentioned "scientific purposes".

This, on the other hand, was for photographic purposes.

He sure would have saved himself some work had he just gotten his hands on a copy of ERDAS, though. And that compressed JPG would look a hell of a lot better as a MrSID image. But I digress.

I still think it's clever. Maybe not groundbreaking or earthshattering, but clever, and neat.

Re:Yar

[Speare]

Well, if you wait a few days for the site to calm down, you can READ his site to get a LOT of information about his processes.

• Use a tripod and a pano head to rotate the camera around the nodal point.
• Take 196 images that overlap slightly in grid formation.
• Use a GUI to assign several control points for each pair of images: image[N]@x1,y1 == image[M]@x2,y2 (He uses his own GUI called PTAssembler for Windows. Others exist, and Hugin works on Linux and Windows.)
• Use an engine to optimize the distortions, and to render the distorted images onto a final image. (He uses Helmut Dersch's panotools, as does PTGui and Hugin and other front-ends.)
• He had additional challenges due to 2GB address limits in Windows (and most 32bit Linux builds would have similar challenges).
• He had additional challenges due to apps breaking with images bigger than 16bit signed coordinate space (and a few Linux tools break on this too).

  • The free-as-in-beer panotools libraries itself is closed-source, and not supported anymore. IPIX(tm) apparently was one of several companies chasing Helmut for patent issues, the resolution of which I am not sure. New work is being done today to open the process up with Open Source equivalents. Otherwise, it's the top tool since it can stitch images taken from any orientation into several projections into several image formats with high quality.

    I use (and help develop) the Hugin tool for my front-end; I've done a few 25 MP images, but nothing so large or as diverse as Max Lyons' works.

How many pixels are enough?

[GCP]

Which makes me wonder how many pixels would be necessary to reach a point where no additional sharpness could be obtained by additional pixels.

The definition in this case is completely filling my field of view (wrap around screen or retinal scanner), allowing me to move my eyes without redrawing, so every point would have to be as sharp as my full center of view (foveal) vision, but without allowing me to move my head (either changing its angle or moving closer to the image).

I can imagine many uses for an even higher resolution image that would allow you to zoom in on interesting spots, but I'm curious about how many pixels the full view scenario above would require. If we just had that, then we could refresh the screen in response to head movements (I wouldn't want to do it for eye movements) and cover pretty much everything, I would think.

Re:How many pixels are enough?

[RhettLivingston]

I can't remember the precise numbers but can approximate fairly easily as long as we change the rules so that I don't have to know what portion of a sphere you can view at full resolution without changing the angle of your head. When browsing photog newsgroups in the past, I found that the generally accepted resolution beyond which a photo to be viewed at a little less than arms length (like holding it in your hand) would not be improved is 170 pixels / inch or 28,900 pixels / sq inch. That is approximately This was based on calculations utilizing the minimum arc that the human eye can distinguish. I just pulled out a tape measure and see that the distance from the approximate center of my head to my hand while holding a picture at a comfortable viewing distance is about 24 inches. The surface area of a sphere with 24" radius is 4*pi*24^2 or 7238 sq inches. At 28,900 pixels / sq inch, that would be 209,168,200 pixels. So, assuming that you must stay at the center of the sphere but that you can look in any direction, this gigapixel photo contains far more resolution than is actually required to meet your specified goals.

More interesting to me would be the answer to a question like, what storage capacity per day is required to capture a full motion, with depth information for every pixel, 360d spherical recording of every moment of ones life with sound, some zoom capacity (I've utilized 35X in my photog experience and would like to see that), and reasonable ability to freeze frame motion of the speeds encountered in everyday life and extract nicely focused still images from that. When someone can either carry storage capacity like that in a pocket sized computer or when the future WIFI equivalent can send that much bandwidth to a home server, our lives will be drastically changed. Roughly calculating this out it comes to about 87 Petabytes / day uncompressed. Compression technology might drop that to 1 PB/day. Should happen about 39 years from now so I'll likely see the day. What a future.

Re:How do you print it?

[xpccx]

I downloaded an archival quality (200Meg) image of a map from the Library of Congress that I wanted printed. Since the Library of Congress charges $200-$300 to print the maps I called the local Kinko's, or maybe it was Sir Speedy. Anyway, I asked them how large of a print they could do and the guy told me the largest he had done was something like 20+ ft by 20+ ft. They printed the map on thick, almost vinyl, paper and it came out to something like 4ft by 3ft. Not only did it look amazing, but it cost me less than $50.

I'd bet any decent frame shop could frame a very large image. I'd guess they'd charge you several hundred dollars for the custom frame though.