Applications of Emergent Phenomena - Systematic Design for Emergence in Cellular Nonlinear Networks - page 131

system is demanded to be portable and integrated within the image sensor. For
instance a blind's aid can benefit of such a sensor attached to the blind's googles.
It can “read” the visual field providing the blind person with a sequence of codes
representing the scene. The codes can be translated in musical tones to be easily
recognized. Other applications are in industry where they can operate similarly to
bar code scanners, but in such circumstances where a bar code is not available. For
instance they can be used to rapidly trace license plate numbers or box labels in a
wireless system.
Fig. 8.1. The monochrome (binary) visual field has to be processed such that a rectangle
will be generated around each isolated character. From such rectangles it is then easy to ex-
tract a list containing coordinates and sizes for each rectangle. The content of the rectangle
is then submitted to a classifier
In [88] we took an original approach and proposed a smart sensor architecture,
which exploits emergent phenomena in simple cellular systems. First we were
looking for the simplest 2D grid cellular systems capable of emergent computa-
tion. Our choice was for CAs belonging to the “2s5” family. Such systems can be
implemented in either digital [66] or mixed-signal [37] VLSI technologies. Par-
ticularly, the mixed-signal technology leads in this case to a very low implementa-
tion complexity. Tools for detecting emergence introduced in Chap. 5 were then
used to discover some “useful” genes among the whole set of 1,024 possibilities.
Surprisingly, it turned out that by using only two genes (ID = 768 and ID = 1,017)
the task of character segmentation can be achieved in about one hundred of CA
iterations. The cell ID = 768 is a linearly separable one, while ID= 1,017 can be
replaced with a slight modification (where an inverter is placed before the central
input of the cell) with ID = 63 such that it can also admit a simple linear threshold
gate implementation. The resulting CA array can process the input image (e.g. the