Simulation 1: Triangle-Square Discrimination

An image contains either a square or a triangle. Their size and position in a frame are random.
Image background(frame) size: 320 by 240(Column by Row) in pixels.
Images have noise. 5 pixels out of 1000 turn its color. white pixels turn black and black pixels turn white.
Square size range: From 20 by 20 (side by side) to 100 by 100, in pixels.
Triangle size range: From 40 by 20 (base by height) to 200 by 100, in pixels.
The number of images per inidividual: 20(10 triangles, 10 squares)
The number of Step: 50(The Retina Window can move 50 times maximum per image)
Fitness function: The fitness is computed only by the number of the step the Retina Window respond correctly, divided by the total number of steps.

Figure 2: Fitness values. Average of the 5 runs. The thick blue line is maximum fitness. The thin green line is average fitness.

The strategy the Retina Window took was as follows. By default, it answers "The image contains a triangle.". Look for the figure. Examine one of its side and see if it is verticle or not. If it is, the Retina Window change its answer to "The image contains a square." and stays at the side. If it is not, the Retina Window move off from the side and look for another figure.
Test of the best individuals of the last generation from the 5 runs of simulations is done with 20 images(10 squares, 10 triangles). 20.66 % of all steps selected averaging filtering method.

Figure 3: Sanpshot from a simulation:
(Left) Examing a side of a triangle. (Right) Examing a side of a square.

Videos
See the animation (slow motion). The Retina Window in Action!!
Retina Window lets us know whether it is answering correctly or not by the color of its window. Green window signfies it is correct and red window incorrect.

Square: AVI file
Triangle: AVI file

Published: 01.03.01 / tk Last update: 22.07.03/jcz (based on 11.06.01/jb)