| OCR Text |
Show CHAPTER 6 PERFORMANCE COMPARISON In the previous chapter, I presented the detailed experimental results o f the adaptive image segmentation system using genetic algorithms. In this chapter, I explore some other search techniques to compare and demonstrate the performance o f the genetic search technique with the other techniques. First, a hybrid search scheme was designed by combining a genetic algorithm and a hill climbing technique and experimented to show the performance improvement over the genetic algorithm alone. Second, the optimization capabilities o f the adaptive image segmentation system were compared with those o f a random walk technique. Lastly, to demonstrate the effectiveness o f the genetic operators, I compared the performance o f the pure genetic algorithm with other variations o f the genetic algorithm in which either the reproduction or crossover operator is eliminated. ' 6.1 Hybrid Search Scheme Genetic algorithms [22, 26] have been proven and shown to provide robust search performance across a broad spectrum o f problems. They provided very promising results for the adaptive image segmentation experiments. However, the hybrid scheme [1] that combines the robust global search technique, genetic algorithm, with the specialized local search technique, hill climbing, can provide performance improvements o ver the genetic algorithm alone by taking advantage o f both the genetic algorithm's global search ability and the hill climbing's local convergence ability. In a sense, the genetic algorithm finds the hills and the hill climber goes and climbs them. This section describes algorithmic details o f the hybrid combina tion o f two techniques and the next sectio n comp a res the |
