Here in HBworld, you'll see one of two ideas pop up from time to time:

1. An infinite library of images/sounds.

2. An infinitely long 'key' that needs only a pointer to release any sound/picture.

To avoid the knee-jerk bones here, I'll start by saying that (1.) is really, really big and mostly full of fuzz and white noise, and that (it is mathematically proved that...) (2.) needs a 'pointer' at least as long as the description of of the picture/song itself!

This idea is about neither, but related to both.

To be Douglas Adams-ish about it, a pleasing picture is one which is almost exactly unlike a screen of random noise, and a pleasing song is almost exactly unlike white noise (unless you're a proponent of the lowercase music scene).

For instance, pick a word a random. Enter that word into google images. Now take one of those pictures at ramdom.

Without seeing your image, I'd warrant it'd be somewhat likely that your image has larger blocks of solid colour in the top 1/3 and bottom 1/3 than in the middle 1/3.

I'd also warrant that only a small subset of the full colour spectrum is likely needed to reproduce your image. Solid areas of colour are slightly more likely to end at an edge which is straight or curved than swirly.

Further, large solid areas of colour are slightly more likely to have lighter hues towards the top.

So, if we can write down as many reasons why the photos we take (or 'like') are UNLIKE pure noise (unless you're into noise, weirdo), we might be able to write a program that contructs images that are somewhat more likely to look like 'something' than 'nothing'.

Note that it would need to be in the CONSTRUCTION of the image that we use these rules. [We can use neural nets to weed out 'noise' images in the 'Ultimate Gallery' but doing so takes longer than the universe has been around.]

Now, in stage 2, the neural nets go to work weeding out images that don't fit a set of pre-trained nets.

In stage 3 we use the internet to get users to 'weed out' even more images based on quick sets of, say 100 in a minute. (Our brains are very quick at 'abstract' vs ' there's something there!')

Stage 4? Well, I'm probably boned out of existance by now. But we might just have an 'ultimate' sound/image archive smaller than 10 ^ 1000000000000 in size?

OK ignore me. :)
-- not_only_but_also, Dec 13 2004

lowercase music http://www.wired.co...tw=wn_story_related
[not_only_but_also, Dec 13 2004]

Komar & Melamid: Most Wanted http://www.diacenter.org/km/painting.html
Surveyed people about what they preferred in a picture, then painted that. [jutta, Dec 13 2004]

Halfbakery: amihotornot breeder amihotornot_20breeder
Similar idea, but for portraits. [jutta, Dec 13 2004]

Soldier, Komar & Melamid: People's Choice Music http://www.diacenter.org/km/musiccd.html
Similar principle to the paintings, but for music. [jutta, Dec 13 2004]

Evolutionary, Genetic, and Organic Art http://snaffle.user.../form/evolutio.html
Maybe one of these people matches what you're imagining. [jutta, Dec 13 2004]

Even as an artist I can't explain what makes a pretty picture , but music is very close to pink noise except that it rings at specific frequencies which are specified by multiplying 440 by twelfths roots of 2 (notes) and doubles, triples, etc of these (harmonics) and rhythm for timing, and of voice, there is the vocal cord vibrations at 100-400 hz frequency, plus 5 resonances which change and determine the vowel sounds, and occasional hisses for consonants or whispers. Chords tend to follow in progressions and are named by the note they sound like, basically. So what I am saying is that normal music has a scientific definition.

However, music videos look quite ugly on average. I suppose that good video is done with good art, psychedelic effects, and synchronizes well with the music, as well as has good looking people singing or dancing or whatever.

You might find this strange coming from me, a Junk Jukebox advocate, but I am also an artist and as a technician I find the "junk" methods to be cheap and efficient, and don't see them lacking in potential for generating excellent output.

To me the real junk is the hard drive you throw away every 3 years, and the gold and platinum is those 100 year old groovy frisbees that all you have to do is spin and put a needle on to get music out of.

Viacom (I think) has a "jukebox" with everything ever recorded or filmed for public viewing on it. If they'd serve that on demand for a reasonable price, we'd all be happy with the media I think.
-- mr2560, Dec 13 2004


Could you restarte what the actual invention is? Your title talks about sound, but your algorithm talks about images.

You want to use genetic algorithms and surveys to investigate which artifacts are most pleasing to the most people?
-- jutta, Dec 13 2004


The idea of of 'The Ultimate Gallery' (see HB idea of the same name) is cool until you realise that every combination of pixels -- even for a very small 80 x 60 image -- produces a huge volume of images, and most of them are of 'nothing'.

Culling 'nothing' from the huge library, even at thousands per second takes more time than the universe has been around.

Perhaps by creating, or 'rendering' the images based on a few rules we could cut down the number of target images from billions of billions to hundreds of billions?

So the idea is: An image generating scheme that works with gradients and blocks of colours to produce every image that is LIKELY to occur by taking pictures of real things, rather than every combination of pixels the CCD can POSSIBLY detect.

I think the scale is still too big though. Let me modify my VisCode script, crunch some actual numbers, and see what I come up with....

:)
-- not_only_but_also, Dec 14 2004


//You want to use genetic algorithms and surveys to investigate which artifacts are most pleasing to the most people?// [Jutta], My Maglev Trans-Am Desperado Train idea, now in one page draft form will address this.
-- mensmaximus, Dec 14 2004


Just a suggestion. I once played with a simple "neural network" program which when taught a limited number of symbols, would morph any subsequent symbols into ones closest to those it "knew". I think this kind of thing is used in OCR. Perhaps a neural network should be shown all kinds of art, and then if it quickly "recognizes" anything new, generated anyhow, it could help sort out the pretty pictures from the ... other stuff.
-- mr2560, Dec 14 2004


I think the thing most unlike white noise is a single frequency sustained tone. I am not sure which would be preferable to listen to. Good music may represent some median value.
-- bungston, Oct 08 2010



random, halfbakery