Applications of Emergent Phenomena - Systematic Design for Emergence in Cellular Nonlinear Networks - page 144

The following procedure is used to generate the CA-based codebook: An initial
“random” binary state (in this example it has a size of 128×128) is applied to a
cellular automaton (CA) with its cells (defined by the ID code, in the example
ID = 203) designed for emergent computation (with a sieve, defined to favor “pink
noise” spatial frequency distributions. After a certain number T of iterations (128
or 256 in the example) a binary pattern containing a wide spectrum of frequencies
is obtained.
Fig. 8.11. The principle of codebook generation using cellular automata
In the following we will call this pattern a codebook since it can be partitioned
in a number of C code-words (each of wxw size) to be used in a vector quantiza-
tion scheme . Unlike traditional codebooks, the CA-generated codebook is entirely
defined by only a few bits of information defining the Cellular Automata ID (in
the above case 203) and the number of iterations T needed until reaching the state
considered as a codebook.
Encryption features (not analyzed in detail here) may be added, by generating
the pseudo-random initial state with another CA with a different gene (ID = 325 in
the above example) starting from a initial state with a small number of pixels
encoding the encryption key as their position within the CA grid.
Besides compactness, the CA-generated codebooks can be optimized much
easier than classic codebooks obtained through learning. No training sets are
needed, while the overall performances of the codec may be easily optimized us-
ing test signals (such as different images) and by simply adjusting a small number