On one end of the spectrum, take some gradients between different colors and alpha blend them together. There's a problem in that colors averaged together tend to average to gray, leaving the center of the image boring. One way to avoid it might be two half gradients which reach a target color at the middle of the image.
Also consider other shapes of gradients, such as circular (cone). Maybe solve the Laplacian.
We would like to have higher frequency components, so on the opposite end of the spectrum is fractal terrain (fractal clouds). Perhaps apply a blurring filter in some locations so that only some parts are high frequency.
Incidentally, I think the diamond-square algorithm can easily generate periodic terrain of size 2^N.
Another way at it would be to directly modify the frequency components of a 2D discrete cosine transform (probably lapped) or wavelet transform.
(Of course, there have been many previous attempts at synthetic images for various purposes.)
No comments :
Post a Comment