Running out of Images
posted by Andrew Bell 
on February 09, 2009 at 06:37 PM
Andrew Bell
Visual Architect : New York
topics: Visualizers, Content, Animation and 3D, Video Production, and Rock and Roll
I was thinking recently, one day we might run out of new images. Let’s take the current standard for high quality images, 1080p hi def video. It’s surprising to realize that that frame contains a finite number of possible images. I thought it would be interesting to figure out just how many, so I wrote a little Python expression to do the math. The total number of pixels is 1920 horizontally x 1080 vertically = 2,073,600 pixels. There are 256 possible intensities of red, green and blue for each pixel, so that’s 2563 = 16,777,216 possible colors. To figure out how many possible images there are, we need to raise the second number to the power of the first, so 16,777,2162,073,600 = 1.5 * 1014,981,180 possible images. That’s a pretty big number – it’s almost fifteen million digits long. Printing it in 10 point Monaco would take over 2,700 pages of paper. Scientists estimate that there are 1080 atoms in the observable universe – a tiny number in comparison.
However big it may be, the fact that the number is finite is a surprising thing to realize. It means that every possible image has a unique ID number. So instead of asking me, “did you see that picture of MIA performing pregnant at the Grammys”, you might ask, “did you see image number 1,394,239,...,572?” Obviously that is totally impractical and it would make you a huge nerd, but it’s interesting that you could.
We think of visual artists as generally creating things that are original, and if they’re doing their job, they are. But I think it’s also surprising to realize that in another sense, they are just exploring a fixed set of possibilities. “After all that hard work and all those revisions, we decided on image number 884,297,...,493.”
Another weird thing is that the answer to nearly any question is contained in these numbers. Who killed JFK? There is a number which is a picture of the answer. What would Michael Phelps hitting a bong wearing an orange floral sari on the moon on July 19, 2033 at 3:19pm look like? What does a McDonald’s look like flooded with water? The answer to that question in full color and high definition is contained in image number whatever. Strange but true – the things which are, the things which are not and the things which will be – they all have a number.
Most of these possible images are pretty similar to what you get if you crank up the Add Noise filter in Photoshop, and aren’t really that interesting to look at. Part of my job at The Barbarian Group is to help figure out technical ways to create images. Put another way, my job is finding efficient ways to skip over the boring numbers and arrive at the interesting ones.
Most of these possible images are pretty similar to what you get if you crank up the Add Noise filter in Photoshop, and aren’t really that interesting to look at. Part of my job at The Barbarian Group is to help figure out technical ways to create images. Put another way, my job is finding efficient ways to skip over the boring numbers and arrive at the interesting ones.
Update:
Thanks to an anonymous commenter who pointed out that I misplaced a comma. Fixed.
Thanks to an anonymous commenter who pointed out that I misplaced a comma. Fixed.
Also, with respect to the successor of an image, let’s assume a straightforward encoding of the image that lines up the pixels row by row, and stores them as interleaved RGB channel ordering. This would make the majority of images’ lower right pixel imperceptibly bluer than its predecessor, but otherwise the two images would be identical.
20 comments
And if there's 'dickcheneyhasahairybutt', there's also 'dickcheneyhasahairybutt AND ONE', but that's where the grain hits in again.
Maybe I'm stretching it too far.
and if a blogging platforms like Blogger or even whatever les Barbarians use had a length limit, could the same be said for posts? which would make this post number 763,629,...,844. or maybe 763,629,...,845. (yea yea, comments too.)
its only 64x64 B&W but still takes pretty damn long to go through all combinations!
Every digital image has a number because every digital image IS a number. Just a very, very, long one.
So your example of "hey, did you see image 1,345,559,...980" is exactly the same as sending someone an email with the image attached, inline.
I think you may be getting stuck on the notion of "numbering" as "ordering". But the set of all possible digital images isn't ordered in any way that would be meaningful for humans. Image 1,345,559,...981 isn't another image of a McDonald's. At best it might be almost but not quite like its immediately preceding image.
It's kind of like a visual type of "bible code" encoding.
You should actually run an evolutionary algorythm that creates random values for the rbg of each pixel and evolved a rules set which produce results that look most like real photos of say out door scenes and then turn the algorythms loose to predict the futre.
Actually if you assumed only a black and white photo of everything you could greatly reduce the variables.
If eachpixel only had a 256 grey value then that help you a lot.
If you can get away with a 24 bit grey value on a 800 x 600 viewport you are REALLY knocking down numbers and could evovle such photos more easily.
vivzizi
vivzizi
You misplaced your comma, it is supposed to be 10 to the 14,981,180th power. The number is really almost 15 million digits long.
I agree with Neil Kandalgaonkar on this one. This seems surprising at first, until you realize that a digital representation of an image is itself an "ID number". I think the reason it seems so surprising is that we have an association with "ID numbers" being short sequences, but in fact in this case the sequence is so long that the number is equally complex as the image.
"evolutionary algorythm that creates random values for the rbg of each pixel"
See the Einstein's Brain Project http://people.ucalgary.ca/~einbrain/new/text/ghosts_in_the_machine.html Where they generate random imagery then use face recognition to pick out the structure.
"there are many implementations of this concept"
I think the best known is "Every Icon" http://www.numeral.com/eicon.html There is another more developed piece from the same year that is fairly unknown, and more similar to the proposition here http://runme.org/project/+godseye/ There are a bunch more documented on http://radicalart.info/AlgorithmicArt/grid/every/ with the earliest one from 1991. That site (radical art) has some great discussion of related concepts as well.