Noise Seeker is suit of algorithms designed for exploring potential order in sensible noise and chaos. More likely than not this project is utterly useless, but the underlying questions raised are nevertheless interesting:
- What are the properties of humanly recognizable 'noise'?
- What are the properties of humanly recognizable patterns?
- What makes an image more recognizable as 'anti-noise', than others?
Given our field of vision there exists an upper boundary for how many 'pixels' we can perceive. If, for the sake of simplicity, we assume that all humans perceive the same amount of 'pixels' and 'ray colors', there must exist a finite number of how many images we can sense with our eyes. Let's assume this number is denoted by X. What percentage of X is recognizable as something of our own world? E.g. lines, circles, or even pictures of humans or houses? Furthermore, can we develop an algorithm to a) detect images that are recognizable by humans as something else than a chaotic canvas, and b) does it exist a pattern of distribution between pictures that are humanly recognizable? Although no parallels can be drawn, a similar question exists for the distribution of prime numbers.
To start off with some sort of benchmark. Consider this randomly generated image:
I'm sure most people will get fairly little enjoyment out of watching that picture. The next image is also generated, but has been through a small fitness function:
The image holds little resemblance to anything in the real world, however we can say that it 'feels less noisy' to look at. It also looks less 'random'. Now for the final generated image:
It certainly looks more recognizable to something unknown than compared to the other two images.
There are some challenges in preventing images gaining fitness points on the basis of pixels that have been 'stacked' together. An analyzer to punish images with stacks of blocks present will be implemented.
