Algorithmic worlds










Search blog posts



Next posts

2009-10-02 Comment gadget

2009-09-23 Algorithmic architecture

2009-09-22 Multiscale art

2009-09-20 Gigapixel picture

2009-09-19 Emergent patterns

2009-09-17 Anselm Reyle

2009-09-09 Non-attenuated pattern piling

2009-09-06 A book about death

2009-09-03 Euler and Twombly

2009-09-01 A new gigapixel picture

Previous posts


A blog about algorithmic art and fractal aesthetic. Click here to subscribe to the RSS feed.

September 22nd 2009

Multiscale art

Except maybe from a few exceptions, visual artworks reproduce and transform visual characteristics of our world to move the spectator and deliver their message. Either straightforwardly in the case of figurative art, or in more twisted and indirect ways. Yet there is a fundamental characteristic of our visual experience that has been completely left out, mainly for practical and technical reasons: the fact that no matter where we look, there is way more structure and information than our eyes and brains can process. When looking at a distant hill, I might see a green spot, but I will fail to see the individual trees forming the forest, the rich texture of their foliage, not to mention the armies of bugs and insects populating it. This is due to the limitations of our visual appartus and of the technology for creating and displaying artworks. The imperfections of our vision are actually what makes it possible for us to see a figurative painting as an accurate depiction of reality, even if objectively it is far from being so.

Recently, a technology allowing to expore this terra incognita of visual art appeared: zoomable digital images. For instance this impressive 17 gigapixel photograph of Yosemite parc gives a more accurate idea of what this landscape really is, even if it is still capturing only an feeble fraction of reality. One may therefore wonder how could visual artists explore this new dimension. In the case of the Yosemite picture, while it beautifully reveals the complexity of our world, the artist's input is meager. There are definitely technical skills needed to produce such a picture, but the artist is not responsible for the details and the complexity appearing in the final image.

Can we imagine painting an image like this one? First, it would have to be painted digitally: a comparable physical painting, assuming milimeter sized details, would be 225x75 square meters... I suspect most digital painters do not use image sizes larger than 2000x2000 pixels, what already outreaches most computer screen resolution. Doing the division, a digital painter who would compose a 17 gigapixel image by painting 2000x2000 pixel tiles would need 4250 of them! Enough to fill a life, or at least a good part of it...

So except from photography, which medium remains? Algorithms! We saw how tedious it would be to paint such pictures by hand, either physically or digitally. But computer-run algorithms are dereasonably efficient at performing tedious work at an amazing speed. Let us be more precise: what conditions should an algorithm satisfy to produce a multiscale image? Of course, now that we contemplate using the full power of algorithms, we do not want to restrict ourselves to images of a given resolution. We would like our algorithm to be able to produce a version of the work at any resolution, provided we let it run long enough. Accordingly, if we can render images at any resolution, let us be purist and require that the work, unlike the mountains of Yosemite (1), displays structures everywhere at every scale. These two requirement have direct implications:

  • The algorithmic work should be a function from some region of the plane (the canvas of the picture) into color space. The algorithm should be a way of approximating numerically this function, the color of each pixel being the approximate numerical value of the function at the corresponding point.
    This ensures that pictures of arbitrarily large resolution of the same work can be produced, by computing more pixels. (2) Luckily, the algorithms which are naturally implemented in Ultra Fractal are precisely of this type.
  • The function that the algorithm computes should be nowhere continuous.
    This requirement ensures that every part of the picture, no matter how small, contains visual structures. If the function was continuous around a point, then at a sufficient magnification around this point, the picture would look monochromatic. We want to exclude this possibility.

Though most of my works are close to this ideal, it turns out that except a few very recent exception, none of them matches the second criterion. Indeed, among the pattern piling algorithms, only non-attenuated pattern piling produces nowhere continuous patterns. I will explain why in a futur post.

20090912-1, an algorithmic work satisfying the two criterions defined in the text.

(1) Indeed, there is a scale at which the notion of "visual structure" cease to make sense. For instance the wavelength of visible light, of the order of 100 nanometers.
(2) Note that this implies that the work is an abstract object (a function) and that only approximations of it can be computed and displayed. It would be interesting to know if this abstract concept of a work already appeared in the contemporary art world.

comments powered by Disqus
Copyright S.Monnier 2009-2021. Bookmark or share