Introduction
According to several recent studies, the self-organization of neurons into brain-like structures is in many respects not unlike the self-organization of ants into a swarm. In other words, perceptive capabilities seem to be able to emerge and evolve from the interaction of many simple local rules, a principle that is now being applied to pattern recognition and its sub-problem of image segmentation.

The new effort takes as its basis the work of Chialvo and Millonas, who developed the first numerical simulation in which swarm cognitive map formation could be explained. From there, an extended model is presented of digital image habitats in which artificial ants react to and perceive the environment in which they are placed. In fact, early results with regard to the evolution of pheromone fields seem to demonstrate that the artificial ant colonies can react and adapt appropriately to any type of digital habitat.

New Twist on an Old Idea
The colony or hive analogy is not new of course. For example, in his well known 1979 book Godel, Escher, Bach, Douglas Hofstadter explores the difference between an ant colony as a whole and the individuals that compose it. According to Hofstadter, the behavior of the whole colony is of a very different character and far more sophisticated than the behavior of the individual ants. In other words, a colony's collective behavior exceeds the sum of its individual member’s actions (so-called emergence), and is most easily observed when studying the ants foraging activity. Most species of ants forage collectively, using chemical recruitment strategies (pheromone trails) to lead their fellow nest-mates to food sources.

The analogy with the way that natural ant colonies work and migrate was also the inspiration of Ant System, a computational paradigm developed by Marco Dorigo in the early 1990s. In Dorigo’s original studies there was no pre-commitment to any particular representational scheme: the desired behavior is specified, but there is minimal explication as to the mechanism required to generate that behavior, i.e. the global behavior evolves from the many relations of multiple simple behaviors.

Since then, several studies have applied this paradigm—or other analogous ones—to real case problems, with successful results. As a heuristic, these techniques have the following desirable characteristics: (1) It is versatile, in that it can be applied to similar versions of the same problem; (2) It is robust, in that it can be applied to other problems with only minimal changes, e.g. combinatorial optimization problems such as the quadratic assignment problem, the traveling salesman problem, and the job-shop scheduling problem; and (3) It is a population based approach. This last property is interesting in that it allows the exploitation of positive feedback as a search mechanism. This is because collective behavior that emerges is a form of autocatalytic "snow ball" effect behavior that reinforces itself—the more ants that follow a trail, the more attractive the trail becomes. It also makes the system amenable to parallel implementations.

At the same time, however, never before have these paradigms been applied to pattern recognition problems or image segmentation, though it seems a natural fit. Segmentation can be viewed as a clustering and combinatorial problem, while the gray level image itself can be viewed as a topographic map, where the image is the ant colony habitat.

Moreover Chialvo and Millonas’ work is an excellent place to start, in that it employs one of the simplest paradigms. Local, memoryless, homogeneous, and isotropic models lead to trail forming, where the trails and networks of ant traffic are not imposed by any special boundary conditions, lattice topology, or additional behavioral rules. Currently the model operates in a finite square lattice. The main goal, however, is to achieve a global perception of one image as the emergent sum of local perceptions of the whole colony, and the model is being extended to digital image lattices, where the respective gray level intensities will be considered in each ant's perception of his neighborhood. Find out more in http://alfa.ist.utl.pt/~cvrm/staff/vramos click on ref. 29. [2000 publications]

More Features