Structural information theory
|
|
Structural information theory (SIT) is a general theory of pattern perception, set about by Emanuel Leeuwenberg in the 1960s. In [1] (http://www.uni-leipzig.de/fechnerday/generalinfo/PDFs/HGeissler) it is argued that SIT is the only approach to Gestalt psychology that spawned a formal calculus that can generate a plausible Gestalt representation of a perceptual phenomenon.
In Peter van der Helm's article "Accessibility, a criterion for regularity and hierarchy in visual pattern codes" (1991) three regularity rules, or compression operators, are proposed: Iteration, Symmetry and Alternation (the ISA codes). They are shown to satisfy the accessibility criteria called transparency and holography. Their perceptual relevance has been experimentally verified at the NICI in Nijmegen.
In 1998 Mehdi Dastani ([2] (http://citeseer.nj.nec.com/dastani98languages.html)) introduced a more formally defined algebra for SIT, using more general operators (besides the ISA codes), and allowing for domain dependent operators (DDOs).
Simplest pattern structure
An important concept in SIT is the simplest pattern structure, or SPS. Here simplest means the one with the lowest information load, where information load is the amount of information present in a pattern.
The goal of SIT is to come with methods that can find the SPS of a pattern in a limited amount of time. One way to avoid the exponential explosion involved in a brute search for the SPS is to use genetic programming.