Abstract: Simulated annealing and parallel genetic algorithm are two helpful methods which can improve the performance of evolutionary algorithm. The paper applies the global optimum performance of evolutionary programming and the hill climbing performance of simulated annealing, and proposes an Otsu dual-threshold value method based on simulated annealing parallel genetic algorithm for medical image segmentation. In the algorithm, evolutions of subgroups are performed among subgroups in parallel, and the hill climbing performance of simulated annealing is utilized, so this algorithm avoids premature convergence of alone group evolutionary process and improves its convergence efficiency. Experiments show that this new algorithm can achieve exact image segmentation and is of better accuracy and stability than parallel genetic algorithm, and its convergence speed is faster than the Otsu dual-threshold value method based on parallel genetic algorithm.

Key words: medical image segmentation, Otsu, parallel genetic algorithm, simulated annealing