Applications of Emergent Phenomena - Systematic Design for Emergence in Cellular Nonlinear Networks - page 133

In order to achieve this goal the controller will increment and decrement the
counters associated with the row and column selectors and will process bit infor-
mation from the cellular array using a one bit bus. Once the starting position

(upper left corner) and the size (vertical and horizontal number of pixels) are
determined, a feature extraction and recognition algorithm will further process the
cropped pixels within each box, extracting seven features to be further classified
and recognized. Pixels are transferred from the CA array via an 1-bit pixel bus.

Among the 2s5 genes the following were found interesting to be used in our
architecture, as exemplified in Fig. 8.3:

Fig. 8.3. Dynamics of the CA for three particular genes from the “edge” type

Genes 792 , and 768 perform a bounded wave propagation such that after a cer-
tain number of steps (proportional to the perimeter of the final square) the CA
generates surrounding rectangles for each isolated group of pixels (grouping of
pixels assume that no more than one white pixel connects any two black pixels).
Note however that gene 792 is not useful since in some cases it generates rectan-
gles with irregular vertical edges. In either case, the number of iterations until a
steady state is reached is proportional to the perimeter of the largest character to

be processed. Since ID=768 a linearly separable gene (see also Table 3.3 in Chap. 3)

it has a simple implementation formula: .

w 2 E D 4