Applications of Emergent Phenomena - Systematic Design for Emergence in Cellular Nonlinear Networks - page 130

moving waves; (b) It might indicate that more measures are needed in addition to
the uncertainty profile, for a finer discrimination among behaviors. For instance an
additional uncertainty profile, obtained with a different initial state profile.
Nevertheless the theory of probabilistic exponents of growth provides a very
efficient tool in locating not only desired emergent behaviors (e.g. by searching
among those with an uncertainty in the expansion area closer but larger to 1) but
also in locating genes with similar behaviors. For the game of life such cells were
also found (see Chap. 7) by looking in the neighborhood of the “game of life” cell
when mapped into a parametrization.
The above suggests that we have now better tools to explore the wide spaces of
the CA cells. In the next we will consider some applicative examples. Genes for
the CA used in these applications were usually found by “sieving” the 2s5 family
(a relatively small family) with some desired behaviors in mind. Among the re-
sulted genes, those found useful for certain signal processing tasks were selected.
Similar genes may be selected from their parametrization neighborhoods or,
equivalently among those having the same or close uncertainty profiles, as deter-
mined using the methods exposed in Chap. 7. It is interesting to note that with
only several genes from the same 2s5 family we were able to solve quite different
problems. Three such applications were next selected as examples. The first is an
application of gene 768 (defined as belonging to the class “edge” in Chap. 5) for
character segmentation, next there is an application of gene 49 (“unstable near
edge” in Chap. 5) in an original method for sound classification and finally there
is a promising and novel VQ compression method (here exemplified for images)
where CA-generated patterns are used as codebooks. The genes used in this case
belong to the category “stable” as introduced in Chap. 5.
8.2 Smart Sensor for Character Recognition
8.2.1 Motivation and General Description
In recent years there is an increased interest in designing functionally complex, yet
portable and low consumption devices. Character segmentation for instance
represents the basic preprocessing stage in any character recognition system.
Given a monochrome image (which may be the result of a scanning process, as in
the standard OCRs, or an image obtained from a video camera, like from a license
plate or a pallet label) the segmentation, as described herein, should result in a list
giving the precise coordinates and the sizes of the rectangular boundaries for each
character in the visual field. In a further processing stage the associated pixels of
each character could be cropped and submitted furthermore to a classifier. In the
end the system should be capable to submit a sequence of codes representing the
identified objects in the visual field (Fig. 8.1). Often such tasks are performed in
software using expensive and power consuming general-purpose computers with
an adequate image sensor [65]. There are however application niches where such a