Random Image Generator

Miscellaneous Topics
#1

I tought of something awesome today, and i want to bring it to life.
Basically, a program that generates all possible 100x100 pixel images (for performance’s sake lets say 32 shades of grey), and only saves the ones that it detected a face/object in.
I wonder what it would spit out… Probably junk, but maybe there would be some gems?
Actually, it would show every possible thing in the universe, in 32 shades of grey, on a 100 x 100 image… right?
What are your thoughts on this, guys?

PS: Im sure there is something similar/identical on the web, but i can’t seem to find it.
PSS: The number of possible combinations would be 32^(100x100)… which is A LOT.

#2

32^(100x100) is a bit more than just “a lot”. Given that there are an estimated only 10^80 something particles in the observable universe.

#3

Just for the fun of it I took liberty to calculate how many years would take to generate all those images at a 1 img/ms.
The number is too big to paste here so: http://pastebin.com/gs1VGCGX

What you could do is limit that image generation using a cellular automatahttp://en.wikipedia.org/wiki/Cellular_automaton.
Also also, there is a video about the formula you want to play with:

_s5RFgd59ao

#4

Instead of asking “is this a face or not” and iterating over all possible images, figure out some way of determining “how much does this look like a face” and then you can make it an optimization problem, which allows for methods of search pruning.

#5

@jonjava

@Lucasmontec thats what i expected tbh xd Also, i saw that video some time ago. It’s nothing special, just a way to encode and decode a b&w bitmap image. I don’t see whats all the fuzz about in the internet. But i must admit, it was fun to write a encoder. I then encoded a funny image, copied the number to my phone, then wrote it in my friend’s notebook ;D 8) (next level d*ck drawing)

@BurntPizza I will try to figure it out on a rainy evening. I’ve never dealt with decision trees before :persecutioncomplex:

#6

You can reverse engineer face detection algorithms to generate face-like images. This narrows your data set and solves the ‘’’’‘searching’’’’’ problem. I’m curious by the way, why face generation like this? Why not just background faces that have certain assets at certain places? Or morphing? Anything would be easier. lol