a digital scan of a 35mm film image of a processing sketch running on an LCD
Skip to Content

Billions and Billions of Photos

Will we ever run out of photographs?

As the web accumulates photos (another 6,037 posted to Flickr in the last minute...), and as projects like Photosynth start to infer intermediate perspectives between these images, one could conceive of collisions — multiple instances the same image, pixel for pixel, from two different authors.

Multi-megapixel full-color pixel images weighing in at dozens of megabytes distance us from what, exactly, these images are at the computationally atomic level of bits. Going back to the same concept on a smaller scale, makes the simplicity of the computational representation of an image clear. Black and white makes this exceedingly basic — an image can be as simple as a grid of 0s and 1s:

Ditch the grid, and you're left with a big binary number:


Convert that to base ten, and you have:

The picture's just a number. It's just one of the possible black and white images that could fit in the 12 x 16 pixel grid. In fact, it's image number 3,138,742,632,065,979,126,417,490,138,422,209,858,437,835,786,299,168,264,319 out of 6,277,101,735,386,680,763,835,789,423,207,666,416,102,355,444,464,034,512,896 possible images.

Every single one of the ~6 octodecillion possible 1 bit 12 x 16 images could be generated computationally. We don't need a camera, a photographer, or an icon artist — a simple program looping through and rendering each possibility would, eventually, stumble upon that exact spray can. And, eventually, it would run out of images to render, and the creative possibilities of a 12 x 16 pixel 1 bit canvas would be exhausted, or, at least, definitively explored.

These numbers are huge — incomprehensible. For all practical intents and purposes, these numbers are as good as infinity, and yet, they're finite in every objective sense.

Scaling this idea up to the high-res color images typical of the web, the number gets even more ridiculous. You can find the number of possible distinct images for a given bit depth and resolution with the following formula:

Possible Photos =(2^Bit Depth)^(Width * Height)

This brings up a few questions.

First, this exposes an inherent and often unconsidered flaw of digital mediums — finite resolution means finite potential representations, suggesting a degree of determinism in digital systems.

Second, what might one discover by traversing the range of possible images with an algorithm instead of a computer? Would anything be discovered? Might we see glimpses of the future or past? By definition, these photographs would all have to be present in the set — portraits of every human, past and present, confidential texts, etc.

Third, what does this say about photographic creativity? Is creating a photo an act of genesis and original thought, or is the photographer merely colliding with an inevitable, preexisting combination of bits?

Fourth, exploring data sets like this also brings up the question: What's the threshold of perception? How many pixels do we need to have before something representational emerges? Of course, deciding on any threshold implies compromise, and the representation of different facets of reality can move this threshold radically. For example, a single black pixel could be said to represent a dark room, and therefore could be said to hold some representational qualities. Representing text on a page or the nuances of a tree, however, would require thousands (maybe millions) of pixels.

A few artists explored this idea in the 90s:
Jim Campbel's The End (1996)
John Simon's Every Icon (1997)

It's also worth considering that this concept is agnostic to the actual way in which in a particular bit string is interpreted. For any digital object of a particular size, this theory of finite possibilities can be applied and explored.

For example, artist Kyle McDonald explored this idea in sound, generating every possible sequence of the chromatic scale's 12 notes:

September 24 2009 at 3 AM

Its an astronomically unthinkable amount of computing power when you start talking about generating something similar to a modern-sized digital photo, especially if it has 48 bit color depth. probably a lot more time needed to compute meaningful results than the lifespan of the universe. Hell, even at web resolution at 256 colors. But i think you are really onto something brilliant here starting with the small grid of binary color.

The binary colored small grid of pixels you describe itself has a pretty immense calculation overhead, but i don't think its outside of current capabilities at all. I feel like the computational overhead could be reduced quite a bit by introducing some form of genetic algorithm into your program (http://en.wikipedia.org/wiki/Genetic_algorithm). You could perhaps train your algorithm from both intuitions of rules you tell it yourself AND by feeding it thousands of small binary grid images that humans recognize for it to study and pattern match.

For instance, using this pattern recognition method I think it would be relatively easy to eliminate most generated pictures that resemble "tv static/snow" by adding rules about the distribution of single black pixels... "avoid checkboard patterns", or even go a step simpler on your generated images and say "all ON pixels cannot be surrounded by OFF pixels (no orphaned black dots)" or defining a minimum ON area tolerance (5 connected ON pixels, 10 connected ON pixels).

I think a great starting point to set your program after is generating organic forms: ferns, veins in hands, lighning bolts, trees, because there's a reason these shapes always turn up in nature. in this way you could start with the bottom center pixel on, and have the program explore only path variations that grow from the starting ON pixel.

I can't recall the documentary name right now, but there's an 80s documentary with Richard Dawkins, where he shows a genetic organic form drawing program that he wrote to use as evidence in studying evolution. Have you seen that? it comes to mind. After he permutated the forms over many thousands of generations, the computer on its own came up with forms that looked just like trees, lightning bolts, etcetera. I will look for it if you are interested.

I think you could simplify your target result set even further this way and maybe reduce the orders of magnitude quite a bit. Hell, if i get time later this year i might nerd out on this. Giving you credit, of course.

Amazing article and site, thanks.

February 17 2011 at 4 PM