|
|
The Project that I worked on involved Evolutionary Algorithms (EA's)
and image processing.
Abstract:
We are designing an object-oriented system which will facilitate the
use of various image processing techniques in conjunction with evolutionary
algorithms . Our main focus is to create a package that employs the power
of evolutionary algorithms while remaining flexible enough to swap in
different image processing techniques such as segmentation or denoising.
The evolutionary algorithm can be applied to any image processing algorithm
with multiple input parameters for optimization. The resulting package
will help in developing an automated system capable of dynamic self-optimizing
computer vision tasks.
Introduction:
Most image processing techniques are either suited only for certain types
of images or they have input parameters that must be adjusted (optimized)
for specific image types. To truly automate a computer vision system,
it would be necessary to develop image-processing algorithms that can
be optimized on the fly. Unfortunately, algorithm’s input parameters
typically interact in a complex and non-linear fashion (Fig. 1). This
makes the task of optimization a difficult one. Having the ability to
automate a generalized image-processing system would make many computer
vision applications much more robust by greatly reducing the required
amount of human interaction with the system to maintain proper operation.
Fig. 1. Finding a global minimum in
a complex, non-linear search space can be very difficult. It is a
perfect application for an evolutionary algorithm. Note that the Phoenix
algorithm supplies a complex, seventeen-dimensional, non-linear search
space.
|
|
 |
|
