9
I Inflexibility in design and system modification (evolutive maintenance)
Therefore, we can formulate two different classes of limitations:
a)
Limitations at the level of software and algorithms for manipulating pixels and
extracting meaningful data
b)
Limitations at the level of hardware especially with regard to the flexibility in
design and re-configurability
We are looking for a specific design and appropriate architecture(s) to cover both of these
limitations.
Cellular Neural Networks
(CNN) do offer parallel processing paradigm that is
robust for image processing. It is a highly parallel architecture with local connectivity
between cells [18, 19]. This makes it a very interesting platform for an implementation on
hardware. In contrast to sequential computing architectures that are highly dependent on
bus bandwidth and width, amount of memory, cache memory size, buffering, processor
clock rate, and CPU [20, 21], CNN is highly scalable and flexible in design.
Research question 3: What is the huge potential of neurocomputing involving either
traditional NN or CNN for high-speed and flexible image
