Mehdi Dastani

Visual Perception

Visual patterns have many interpretations. Human perceivers prefer in most cases one of these interpretations. For example, the visual patterns illustrated in Figure 1 and Figure 2 may have, among others, the illustrated interpretations. In these two cases, human perceivers usually prefer the interpretations that are illustrated by the red color.

Contact: Mehdi Dastani

Structural Information Theory

Structural Information Theory (SIT) is a general theory of pattern perception which aims at explaining why human perceivers prefer certain interpretations of patterns rather than arbitrary ones. The explanation is based on the assumption that the human perceptual system is sensitive to certain types of structural regularities of patterns called iteration, symmetry, and alternation. The relevant SIT papers are as follows:

Leeuwenberg, E. (1971). A perceptual coding language for visual and auditory patterns. American Journal of Psychology, 84, 307-349.
Van der Helm, P. and Leeuwenberg, E. (1991). Accessibility: A criterion for regularity and hierarchy in visual pattern code. Journal of Mathematical Psychology, 35, 151-213.
Van der Helm, P. (1994). The dynamics of Prägnanz. Psychological Research, 56, 224-236.
Dastani, M. (1998). PhD Thesis, University of Amsterdam, The Netherlands.
M. Dastani, Elena Marchiori, and Robert Voorn, Finding Perceived Pattern Structures using Genetic Programming. In the proceedings of Genetic and Evolutionary Computation Conference (GECCO-2001).
Robert Voorn, Computing Perceptual Structures of Visual Patterns Using Genetic Algorithm, Master Thesis, August 2001.

Implementations

This implementation is the result of the Master thesis of Robert Voorn and it is designed to determine the perceptual structures of string patterns based on genetic algorithm. The system is implemented in three versions: NoUnit, Unit, DDOext. The differences between these versions are explained in the master thesis of Robert Voorn. For each version a SubDirectory is made containing the source code of two different GA's: Generational + roulette-wheel selection and Steady state + tournament selection. The input file should contain a string the perceptual structure of which should be determined.

NoUnit.zip: the basic ISA operators are covered, but not the unit operator
Unit.zip: the basic ISA operators extended with the Unit operator
DDOext.zip: the alternation and iteration operators of the Unit version are modified to cover domain-specific regularities as well

In the DATATYPEGP.CPP file four different compiler-directives are present.

DEBUG : can be used to turn on extra checking functions that are useful for debugging the program. Also extra output is given.
S1 - S3 : these are three different Solutions used in the thesis and can be switched on and off seperately.

In order to run each of these versions (under unix), download the corresponding directory, execute ./make in that directory. A file is then generated called GenGa. You can now run the program by typing the following command on the unix prompt:

./GenGA pCross pMut pComp generations poolsize inputfile runs

The arguments of GenGA are as follows:

pCross : crossover rate (0..1)
pMut : mutation rate (0..1)
pComp : weight for adding nr of operator to fitness (0..1)
generations : amount of generations
poolsize : size of the pool
inputfile : this file will be read for the string to be used in the GA
runs : this the amount of runs to be performed

Last modified by Mehdi Dastani on September 2002.

Structural Information Theory

Implementations

Related Sites