Interactive Evolutionary Algorithms for Image Enhancement and Creation

Image enhancement and creation, particularly for aesthetic purposes, are tasks for which the use of interactive evolutionary algorithms would seem to be well suited. Previous work has concentrated on the development of various aspects of the interactive evolutionary algorithms and their application to various image enhancement and creation problems. Robust evaluation of algorithmic design options in interactive evolutionary algorithms and the comparison of interactive evolutionary algorithms to alternative approaches to achieving the same goals is generally less well addressed.

The work presented in this thesis is primarily concerned with different interactive evolutionary algorithms, search spaces, and operators for setting the input values required by image processing and image creation tasks. A secondary concern is determining when the use of the interactive evolutionary algorithm approach to image enhancement problems is warranted and how it compares with alternative approaches. Various interactive evolutionary algorithms were implemented and compared in a number of specifically devised experiments using tasks of varying complexity. A novel aspect of this thesis, with regards to other work in the study of interactive evolutionary algorithms, was that statistical analysis of the data gathered from the experiments was performed. This analysis demonstrated, contrary to popular assumption, that the choice of algorithm parameters, operators, search spaces, and even the underlying evolutionary algorithm has little effect on the quality of the resulting images or the time it takes to develop them. It was found that the interaction methods chosen when implementing the user interface of the interactive evolutionary algorithms had a greater influence on the performances of the algorithms.