So a program which constrains our drawing choices into a particular
style could help
Even more help would be a program which generates random art from
which we could chose the ones we like
This lecture is based on my own research
Slide 4 : 4/21: Piet Mondrian
Piet Mondrian
b. 1872 Netherlands, d. 1944 New York
Contributor to the De Stijl art movement
Attempts at mathematical analysis of his compositions
E.g. some of Mondrian's compositions are triple connected, in that you
can not separate the graph into two without cutting at least three
lines. This applies to about half of Mondrian's work in the period
1918 - 1938
Slide 5 : 5/21: Is it Art?
Is it Art?
Mondrian's work appears to be simple compositions
Lee (2001): art students could not identify genuine Mondrian compositions
But McManus (et al, 1993) found majority of subjects could distinguish
between original and modified
Wolach (2005) found subjects preferred Mondrian's line spacings
=> general aesthetics?
BUT if they could chose own preference, then the preference for
Mondrian spacings vanish
=> individual aesthetics?
Slide 6 : 6/21: Construct images
Constructing Mondrian-like images
Random initial points
Imaginary lines
Some imaginary lines selected
Skeleton filled in
Colour(s) added
1.
2.
3.
4.
Slide 7 : 7/21: Evolutionary Algorithms
Evolutionary Algorithms
Programs which solve 'problems' by simulating Darwinian selection
among solutions
Individuals are assessed for fitness and create 'offspring'
Over time, accumulate good components: better solutions
Randomness in creating images = minor creativity (or at least an
illusion of it)
Slide 8 : 8/21: Artificial AI
Artificial AI ¿=? Crowdsourcing
How to evaluate art generated by our program?????
AI = Artificial Intelligence => machine helps human
So artificial AI => human helps machine!
Similar to "crowdsourcing"
Our work: user selects the images s/he likes
We generate more images
Over time we generate 'better' images.
Better for a _particular_ user!
Slide 9 : 9/21: Interface
User interface & process
Slide 10 : 10/21: Example of Use
An example of use of GUI
Slide 11 : 11/21: Results Z
Results - subject Z
Chose 20 from 2000
Red + Blue
Mostly touch along edge
Slide 12 : 12/21: Results J
Results - subject J
Small blocks (mostly)
Close to an edge
Likes Red
???
Slide 13 : 13/21: Results T
Results - subject T
Medium sized blocks
Some space between blocks
Context?
11 Yellow
14 Red
14 Blue
Slide 14 : 14/21: Results B
Results - subject B
I can't see a pattern
Subject couldn't explain it either
Maybe likes multiple kinds of images?
Slide 15 : 15/21: Computer Eval
Computer Evaluation of Art
Trained a neural network using the features of images
2 Longest parallel non-touching coloured areas; Dist. between
Slide 16 : 16/21: NN Results
Neural network results
Subjectivity of experiment unavoidable
Also, length of time experiment takes can change subject preferences
Used only 1 subject (T), for an extended period, evaluating some 5,000
images.
Analysed significance of inputs
The 8 most important inputs account for 50% of the behaviour:
proportion of image which is coloured
biggest horizontal distance between lines
Manhattan distance between 2 colours
min. Euclid. dist. bet. yellow &
blue
proportion of image which is yellow
smallest vertical distance to an edge
min. Euclid. dist. bet. yellow &
red
length 2
colours
touch
Slide 17 : 17/21: NN results T
Neural network results - subject T
unexpected: prop. of image which is yellow???
min. dist. between yellow & blue
min. dist. between yellow & red
11 Yellow
14 Red
14 Blue
Slide 18 : 18/21: Other uses
Use in Design? (mockup only)
GUI field structure would have many parts for an expert,
hide some for laymen
Slide 19 : 19/21: Really Mondrian
Which are really by Mondrian?
(OR, pick three which match somehow)
A
B
1
2
3
Slide 20 : 20/21: Creativity
Creativity
Where is the creativity? Human or computer?
(Does it matter?)
Another tool: MondrianDrawer
Drawing Image mode
This interface allows full use of your creativity in drawing Mondrian-like
images. The program limits you to drawing legal Mondrian images. There are
a number of user settable parameters.
Generated Image mode
This mode is similar to the Darwindrian program in some ways. Here, you
can chose the origin points, and then repeatedly generate Mondrian-like
images for those points. You can convert such an image to the Drawing
mode and modify it manually.
We will be using the MondrianDrawer program in the labs.
Slide 21 : ToC : Art
Table of Contents (21 slides) for the presentation :
