Generative Creativity - lecture 13:
Images

Introduction

Production of images and visual artefacts form another major application area for GC.

Approaches tend to be be more computationally challenging and results are often (though not always) viewed disparagingly.

The question of evaluation becomes particularly challenging.

Cave art?

Cave paintings from Chauvet cave dated to approx 32,000 years ago.

Lions hunting bison.

Horse heads
Note `generative' use of rock-shapes, cracks, lighting effects etc.

14th century tiling

14th Century tile pattern from Nasrid Palaces, Granada, Spain (photo from Roman Verostko)

Algorithmic propagation of basic shapes in interlocking patterns.

Schmidhuber low-complexity art

Schmidhuber's `low complexity art' is based on minimisation of algorithmic-complexity, i.e., minimisation of the shortest possible description for an image.

Schmidhuber's idea is that the subjectively more beautiful image is the one with the simplest (shortest) description, given the observer's method for encoding and memorizing it.

On this assumption, the appreciation of visual beauty works the same was as it does for a mathematician appreciating the beauty of a simple proof.

Schmidhuber butterfly
This image of a butterfly, sunflower and vase (at least I think that's what it is...) was generated by finding the simplest way of constructing the desired shapes out of the available shape primitives.

Fractal art

Image generation using recursive but not necessarily mathematical specification.

Simple tree generation:

  tree = branch branch
  branch = tree
Generates

        /\
       /  \
      /    \
     /      \
    /\      /\
   /  \    /  \
  /\  /\  /\  /\
etc.

Pythagoras tree

Slightly more complex tree generation.

At each stage the drawing of a square gives rise to two smaller squares so that the squares bound a right triangle.

In a stochastic version, right triangles are randomly varied at each stage for a less symmetrical result.

PythagorasTree applet

Sierpinski triangle

Hat fractal

Fractal gallery

Large numbers of `fractal art galleries' available on the web, e.g., http://fractalarts.com/ASF/galleries.html

Fractal generator applets

Real-time fractal generators show how increasing levels of detail are filled-out as the image is zoomed.

Harold Cohen and AARON

Harold Cohen has pursued at 30 year project to model his own creative processes using a rule-based drawing/painting program called AARON.

A well-known early work from his `people on balls' period.

Rules

Rules in AARON cover things such as

Getting the human feel

Cohen has been particularly successful in producing outputs with a `human feel'.

This seems to have been achieved by having the program make use of specifically `human' rather than `computational' methods.

For example, in an early version of the program, Cohen had the program generate a shape (e.g., a leg) by drawing in a few guide lines and then encircling them with a bounary. (These guide lines are visible in some of the pictures featured in Boden's Creative Mind book.)

The AARON demo

The AARON of the demo version uses contour-influenced shading processes (e.g., the texture of coloration is influenced by nearby boundaries).

It also uses various stochastic methods to give the drawing a slightly `random' feel. This includes random selection of colors from tightly defined ranges.

Cohen recently visited Sussex and gave a talk about his most recent work with AARON.

This involves zoomed-in drawings (just the foliage now) and use of random color schemes generated using class-based assignments of color bases and randomisation using HSV (rather than RGB) representation.

What makes artist-models controversial?

GC approaches which use an explicit artist simulation tend to be highly controversial.

These systems seem particularly to threaten our notions about the special qualities of human creativity.

But what's the problem?

Summary

Exercises

Reading

Page created on: Wed Feb 17 10:57:05 GMT 2010
Feedback to Chris Thornton