The Evolution of Stylized Images using Genetic Algorithms

by Steven Bergen and Brian J. Ross

Texture examples:


(a)

(b) (See larger version below).
Texture examples made with genetic programming.
(More information here).


Mona:


(a) Target

(b) Mask

(c) RGB mode, 600 circles
(size 1-40).

(d) Quantized: 128 colours.

(e) Quantized mask:
150 circles (size 1-5)

(f) Mixed: 500 lines (width 1-5),
400 circles (size 1-20).
GA parameters: population size 170, 1000 generations, crossover rate 100%,
mutation rate 75%, tournament size 4, 2-point crossover.


Intervals:


(a) Generation 1

(b) Generation 150

(c) Generation 300

(d) Generation 600

(e) Generation 900

(f) Target
GA parameters: Base chromosome 600 circles (size 10-40), mask chromosome 250 circles (size 1-10),
128 fixed-colour palette, population size 160, crossover rate 100%, mutation rate 80%,
tournament size 4, 2-point crossover.


Red:


(a) Target

(b) Base 400 circles (size 10-40),
mask 100 circles (size 1-10).

(c) 400 lines (width 1-7),
2200 gens.

(d) Circles and rectangles,
sizes 1-40, 3300 gens.

(e) 100 triangles, no size limits,
no alpha, 14200 gens.

(f) 1800 dispersed small circles,
size 8, 5700 gens.

6 colours
GA parameters: population size 170, crossover rate 100%, mutation rate 75%
tournament size 4, 2-point crossover, colour palette 128 quantized
colours, alpha-transparency (unless stated otherwise).


Girl:


(a) Target

(b) Result

(c) Target blurred
(Gaussian, radius=9.2)

(d) Result blurred
(Gaussian, radius=9.2)
GA parameters: base 800 rectangles (size 5-30), mask 250 circles (size 1-5)
population size 168, generations 3000, crossover rate 100%, mutation rate 75%,
tournament size 4, 2-point crossover, 128 quantized colours, alpha.


Painting:


Painting by S. Bergen


Geometric Impressionism:


Portrait by B.J. Ross


Gold:


Target image evolved by genetic programming.


JNetic result

Difference image

Final result after colour editing: Gold by B.J. Ross


Generated images copyright (c) 2009 Steve Bergen and Brian Ross.

Back up: http://www.cosc.brocku.ca/~bross/JNetic/