Greg Santucci's honours project weblog
Weblog for the Automatic Art Generation project
Supervisors:
Alan Dorin (website)
Jon McCormack (website)
About the site:
This weblog is intended to preserve some of the references I accumulate and work I do as I work on this project.
This project is about automating the process of generating interesting images.
Sunday 14th November
Thesis is available for download here.
The source code for the drawing program is available for download here.
You need Qt to compile it. To compile it, run qmake to generate the Makefile, then run make.
Alternatively, I compiled a win32 version for your convenience available for download here.
Here are some images the program has produced:
A zipfile containing many more screenshots and images is available for download here.
Monday 18th October
Artificial life: interesting art,
interesting creatures
Reading:
“Explorations in Art and Technology”
This book is concerned with the introduction of computers into the art world.
Computers have been used in art almost as long as computers themselves have existed.
In 1963, “Computers and Automation” magazine began running an annual competition on computer art.
Max Bense: Shannon and Weaver's concept of information defined the background for attempts by Bense and his students to invent quantitative measures of the aesthetics of objects.
“Generating works of art (aesthetic objects) is as rational an activity as anything in the natural sciences”
When Georg Nees exhibited his computer generated line drawings. Short descriptions of the images appeared in the brochure that described the “algorithmic essence” of the drawings. It was entitled “Projects of generative aesthetics” Some artists who were present became “nervous, hostile, furious”:
If they were done by a computer, how could they possibly be art?
Where was the inspiration, the intuition, the creative act?
What could be the message of these pictures?
“They were nothing but straight lines on white paper, combined into simple geometric shapes. Variations, combinatorics, randomness....but even randomness, the artists learned, was not really random, but only calculated pseudo-randomness, the type of randomness possible on a digital computer. A fake, from start to end, christened as art!”
Max Bense invented the term “Artificial Art” to distinguish the computer products from human (made) pictures.
Two more exhibitions followed in 1965: Frieder Nake in Stuttgart (included some of Nees' work) and Michael Noll and Bela Julesz at the Howard Wise gallery in New York. Noll was awarded 1st prize of the computer art contest of “Computers and Automation” in August of 1965. 1966, Frieder Nake won it.
Jasia Reichardt began the show “Cybernetic Serendipity” Institute of Contemporary Art in London in 1968.
Georg Nees published the first PhD on computer art in 1969. Max Bense was the advisor.
This book is concerned with the interaction between the artist and the “technologist”.
www.creativityandcognition.com
"The Emerging Mind"
Professor Ramachandran's suggested 10 universal laws of art:
“An Anthology of Aesthetic Theory”
Goodman proposes that art objects exhibit 3 symptoms:
Repleteness: For art objects, more of their physical characteristics become relevant to note. For example, the quality of a line is not important to note for a diagram.
Expression: Art objects express things that are not literally part of the work. For example, a “fiery” painting is not literally hot to touch.
Composition: Every part of an art objects makes a meaningful contribution to the whole.
“Cognitive processes in the perception of art”
Various studies were run to find out if
the message of abstract art pieces was getting through to the
audience.
Alternative to Birkhoff's aesthetic measure. Instead of M = O / C, M =
O * C was found to result in a better fit.
"Cognition
and the visual arts"
perceptual organisation, propsensity to group similar things together
tendency to group things that are close to each other
figure and ground
conflicting organisational patterns
aha! conflicting organisational patterns tend to occur with the images generated by my program. My program generates images with arbitrary degrees of rotational symmetry. I have found that one degree of rotational symmetry is a little too boring, 2 is best, 3 is too messy.
Rotational symmetry results in interesting images quite often, because the images exhibit conflicting organisational patterns.
weak and strong forms
competing and stable forms
ambiguous direction of a field of triangles
perceived brightness and context
EYE MOVEMENTS
eye movements reveal how we try to find a structure when there is no question
eye movements across balanced and unbalanced images
perspective, lines that are hidden in the image, directions for the eye to follow
“The Creative Computer”
Two methods of working with the computer. Using a software tool to assist in the creation of art where the end result is known, or by writing a program which can originate artwork itself, end result is unknown.
Classical artists also were affected by their medium, the explorations that were possible. Computers enable painting to become an exploration, according to David Em.
David Em used the facilities at the Jet Propulsion Laboratory to compose abstract “paintings” of startling originality and vividness, of a dreamlike, almost nightmarish quality. David Em manually directs the drawing, but uses the program apply textures, projections, and other geometrical operations.
Drawing by expert system: Harold Cohen's Aaron: randomly places, next random placement constrained, each subsequent random placement more and more constrained.
Running a simulation to compose an image: Chris Briscoe, point masses applying gravitational forces to each other.
Dominic Boreham uses the computer to paint through an iterative process, trying out something, judging it, however he makes the judgments and maintains that the judgments can't be embodied in a computer.
No mention of Roman Verostko, or the “Algorists”.
Negroponte on art and computers: “Rarely have two forces joined forces seemingly to bring out the worst in each other as have computers and art”.
Bad reputation computer art has acquired, pg 162: “Traditionally the computer has been used to produce what you might call 'pretty pictures'. In a sense its a kind of chocolate box art of information technology, which in some ways has given computing in art a bad name, because people have produced random number squiggles and spirals and so on – maybe they have been mathematicians or computer scientists. They've said “Hay! This is just like art!” and put it on their walls and so on. This drives artists and designers mad because if you had done it with a pencil or with string on nails or whatever, people wouldn't look at it twice. Certainly they wouldn't consider it worthy of having an exhibition.”
“the problem as far as the layman is concerned is that most computer art is highly abstract, and abstract art is hard enough form hom to appreciate and judge, superficially or deeply, even when there is no computer involved. Then the aspect of computer creativity is introduced to complicate matters yet further. It is often pointed out that the pictures produced by Harold Cohen's program look very much like the ones he used to paint himself.
...if Jackson Pollock could throw paint at his canvases, the computer might as well too, but the rhyme or reason can seem elusive.”
Cohen: “Primarily, I believe its function is to produce proliferation of the decision space without requiring the artist to “invent” constantly”
But surely the whole business of an artist is to invent!
Negroponte: The technology is the whole point, the medium is the message.
Alan Sutcliffe: Computer art not given a fair chance to prove itself. All art involves the production of a great deal of mediocre art before anything exceptional comes along, and computer art simply needs a longer trial.
Negroponte sees 2 directions: one is works that “get to know” the future owners, as the future owners “get to know” the work. The other direction is computers used for personal art, as opposed to public art. All the things we treasure as individuals, but are meaningless to anyone else: a drawing by one of our children, stone brought back from a holiday. Computers provide a new opportunity for self expression.
“Think of our Sunday painter reincarnated with an easel of electronics and a palette of computer graphics. His work is as invigorating as a game of tennis, his challenge is that of chess, his product is as ephemeral as a child's drawing. In this fantasy lies the potential for the major impact of computers on the visual arts of the future.”
The authors conclude that Cohen's expert system, artist as programmer rather than as software user, is the direction computer art will move in.
Saturday 26th June
Reading:
“Fluid concepts and creative analogies”, Douglas Hofstadter
Scha, Bod (1993): “Computational Aesthetics”
“The Conscious Mind”, David Chalmers
“Kinetic Art”, Guy Brett
“Algorithmic Aesthetic: Computer models for criticism and design in the arts”, George Stiny and James Gips (111.85 STI Gippsland)
“Creative Computer: Machine intelligence and human knowledge”, Donald Michie and Rory Johnston (303.4834 M624 Peninsula)
“Yellow House”, Joanna Mendelssohn (707.4099441 Y43 Matheson)
“Art and its significance: an anthology of aesthetic theory”, compilation edited by Stephen Ross (707.17 R826A Matheson)
"Mathematical Basis of the Arts", Joseph Schillinger (701.17 S334M Matheson)
“Philosophical aesthetics”, Oswald Hanfling (700.1 H238P Matheson)
“Robots: fact, fiction and prediction”, Jasia Reichardt (629.892 REI Caulfield)
Schneider (1995): “A Beginner's Guide to Constructing the Universe”
Reading Notes:
On “Mathematical Basis of the Arts”:
Pages 224 – 230 contain a variety of designs traced out by a moving object which changes direction rhythmically
Pages 284 - 343 contain designs created using only groups of arcs
Visual depictions of rhythm
A variety of designs from simple rules
On “Fluid concepts and creative analogies”:
Describes a system called “copycat” which models cognitive behaviour
A simple domain is chosen to demonstrate the model
The model consists of a workspace, slipnet and coderack
The slipnet is a concept graph where every concept in the domain is represented as a node.
Every arc is associated with a concept.
Arcs are strengthened based on the activation level of corresponding concept nodes, encouraging “slipping” of activations.
The coderack contains the codelets, which are programs waiting to transform or add constraints to the workspace. These are spawned by the active concept nodes, each of which has a reportoire of codelets it generates when active.
The workspace is where the agent adds to and modifies its creation or proposed solution to a problem.
There is a global temperature which regulates how readily concepts gain and lose their activation. The temperature is kept constant or reduced over time, similar to simulated annealing.
I believe this cognitive model can be applied to the “generate interesting images” problem.
Programs:
I have written a program which traces the paths of simple vehicles. By placing them randomly in an area and causing them to follow a random other vehicle, the system generates an image with curvy lines which has a solid curvy feature towards the centre.
Friday 26th March
Reading Notes:
NK Humphrey "The illusion of beauty"
Concerns itself with what could be the evolutionary reason we receive pleasure from experiencing beautiful things.
A beautiful thing exhibits "likeness tempered with difference", like rhyme in poetry.
"What is the biological advantage of seeking out rhyming elements in the environment?"
"Considered as a biological phenomenon, aesthetic preferences stem from a predisposition among animals and men to seek out experiences through which they may learn to classify the objects in the world about them."
"Beautiful structures in nature or art are those which facilitate the task of classification by presenting evidence of the taxonomic relations between things in a way which is informative and easy to grasp."
After describing how Sonata form presents a theme and develops it in such a way that its interesting to follow, but the listener is never lost, Humphrey says this:
"If psychologists could learn from music how best to present evidence of the relations between things in a way that people will find easy to take in, it is equally possible that laymen might learn to do it for themselves. In that way, the work of art would achieve a new importance, as a model of the way we should structure our experience whenever we need to acquire genuinely useful information"
"...through the experience of beauty in works of art we learn to learn"
New stimulus offers potential opportunity of improving internal classification system, so animals and humans are attracted to it.
Perhaps described in information terms, a beautiful thing is something which adds usable information to an information gathering / classifying system.
Neural networks spring to mind...
A neural network begins in a random, 0 information state. After a training process, the neural network has experienced something "beautiful" which has added usable information to the system.
Beauty as entropy?
The drawing program should exhibit the "variations on a theme" concept.
Relationships between drawn objects...
Wednesday 24th March
Reading:
"Design And Form", Johannes Itten (707 I91 DE, Main Library)
"Vehicles: Experiments in Synthetic Psychology", Valentino Braitenberg (152 B814V, Main Library)
"Point And Line To Plane", Wassily Kandinsky (709.47 K16P/d, Main Library)
Humphrey, NK: 1973, The Illusion of Beauty, Perception 2: 429-439. (paper available online)
Websites:
On the GroupC website are examples of computer generated artwork based on the motion of Braitenberg vehicles, as well as some prototypes for interesting "intelligent" UI drawing tool ideas (eg this applet)
Present activity:
Writing a program that draws a set number of lines and points on a plane.
Compare using uniform probability distribution with using other probability distributions (poisson etc)
Look at various ways of describing a line. eg
Two vertices
Origin, orientation and length
Length and two circles line endpoints must sit on
Length and two lines line endpoints must sit on
Other constraints
Establish relationships between drawn objects eg
Having drawn one line, modify some of the line parameters to draw the next line
Sets of lines
Hierarchical
Other heuristics/models controllable by a set of parameters.
If you want to be able to tune the parameters using the Blind Watchmaker technique (website), its important to try to construct the heuristics/model so that small changes in the parameters produce small changes in the drawing. An iterative physical simulation of something is a dynamical model, and these tend to be very sensitive to small changes to the parameters or initial conditions, whereas formulas tend not to be.
I'm aiming to write one drawing program at least every two weeks in order to prototype different approaches to computer generated art, and to have lots of examples to refer to later.
Other miscellaneous web resources:
Karl Sims Page: http://www.genarts.com/karl/
Craig Reynolds, "Steering Behaviors for Autonomous Characters": http://www.red3d.com/cwr/steer/
Blind Watchmaker Website: http://physics.syr.edu/courses/mirror/biomorph/
Geometry Junkyard: http://www.ics.uci.edu/~eppstein/junkyard/
Miscellaneous Geometry functions: http://astronomy.swin.edu.au/~pbourke/geometry/
Game development resources at FlipCode: http://www.flipcode.com
GameAI: www.gameai.com
Review of the book I'm reading: http://cscs.umich.edu/~crshalizi/reviews/vehicles/
More than you can poke a stick at on google
http://people.cs.uchicago.edu/~wiseman/vehicles/