add more burden on the segmentation methods.
In this case the method must beadaptive to these changes and this how it has been approached in this chapter wherethe metaheuristic Genetic Algorithm has been modified to be Hybrid by includingthe Hill-Climbing method. This inclusion reduces the speed and slows down theprocess of image segmentation which can be noticed when increasing the satelliteimage size. To speed up the new modified Genetic Algorithm a new term is introducedthe dynamic population or variable chromosomes where in this new methodthe size of the chromosome is variable. In addition, part of the image is used inrandom process of creation of the population which means that the required timefor reproduction , replacement and fitness calculation will be reduced significantly.Although the accuracy obtained can be judged as a good one, but sometime is nothigher than the one obtained using Hybrid GA alone. In addition, in the HyDyGAthere are several issues which must be checked every time reproduction or mutationis performed which is the differences in the size of the chromosomes and the position(s) of point(s) of selection in the crossover process.
This is a very complex taskwhich is performed successfully. The variability of the population is another issuethat is taken must be taken care at the beginning when creating the population andin the evolution of new population during the running of HyDyGA. One importanttopic which may improve the segmentation process, but may increase the complexityof the metaheuristic Genetic Algorithm is the use of multi-objective GA. Thistopic is not tackled in this chapter due to the lack of enough experiments and workon this issue. However, it is worth listing some of these limited studies such as theone in 68-70. These studies deals with image segmentation problems as problemshaving multiple objectives. This property can be defined as minimizing the distancesbetween objects in the same cluster (intra cluster), and maximizing distancesbetween different clusters (inter cluster). Working with multiple objectives is consideredas a difficult problem, but sometime a multi-objective optimization approachfor some problems is the only suitable method to find a solution 71.
Working withmulti-objective GA adds more burden on the computer resources. Multi-objectiveGA requires that the final result which is the best approximation of the Pareto frontbe considered as multi global optimum segmentation solution. On the other handthis problem is solved by using the combination of metaheuristic algorithm such asGA and another clustering algorithm such as Self-Organizing Maps (SOMs) whichis the case of SOMs-HyGA or the combination of GA with Fuzzy C-Means (FCM).In that case these processes can be run with different settings and can be evaluatedto obtain the best global optimal solution.