39
Chapter 5
5.
CNN template calculation schemes with a particular focus
on the learning/training based approach through Genetic
Algorithms
In this chapter the focus lies on the following research question:
“
What are the major
template calculation schemes of relevance for CNN based image processing? How can these
calculations be performed in a real-
time high performance computing context?”
5.1
Introduction
Cellular Neural Networks technology provides a very
powerful analog computing
architecture for a variety of array computation and image processing tasks [94]. From a
theoretical point of view CNN model offers a huge potential for modeling image processing
filters and operators on a CNN Universal Machine. Each CNN
processor matrix used in
image processing has a feedback template, a feed-forward template and a bias template.
These three templates can reconfigure the CNN model without any changes in hardware.
The most challenging issue is to find an optimum set of proper
template values for each
specific application[98]. Figure 5-1, shows this CNN architecture. Overall, there are three
major ways to calculate the feed-forward and feedback templates:
(a) The Intuitive method:
This first method is needs intuitive thinking of the designer [99]. Depending on the
designer’s experience in either processing
images or dynamics of arrays, we can have a
template. There is no guarantee to find a template for all image processing operators and
could be very difficult to find a template for complex solution.
Experts are familiar with
template of basic image processing operators and they can combine different templates or
performing them on CNN individually one by one. For example, we know the template of
Laplacian of Gaussian for finding edges and also template for smoothing image. If we
40
combine these two templates on a control template and
feedback template respectively,
results will be an enhanced image with sharpen edges.
