A Study On Image Segmentation By An Improved Genetic Algorithm

The image segmentation is the key step from the image processing to the image analysis, and it is also an important and difficult task of computer vision technology. Tough it has been being attached much attention by people for many years, it develops very slowly. Nowadays, many algorithms which are all particular for some special problems have been put forward, but none of them is suitable for all images. It is very difficult for us to find an algorithm which can quickly search, exactly orient and optimize the segmentation in the areas which interest us, especially the algorithm for grey image segmentation based on threshold.Genetic algorithm (GA) is a sort of efficient, parallel, global search method with its inherent virtues of robustness, parallel and self-adaptive characters. It is suitable for searching the optimization result in the large search space. Now it has been applied widely and perfectly in many study fields and engineering areas. In computer vision field GA is increasingly attached more importance. It provides the image segmentation a new and effective method.Algorithms and analyses about image segmentation are presented .An overview on the theories and the recent development is given. Also the status of GA applied in the image segmentation field is presented and the theories, steps, results and analyses of several GA applied in the image segmentation are given.Through the deep research and compartions on the GA fields, an improved self adaptive genetic algorithm is studied for grey image and noisy images. In the new algorithms: coding with 2-dimention chromosome is adopted; initialization of population with stochastic and symmetrical methods is produced to keep the variety of the population; A new rule is imported in the crossover operation to avoid the population degenerating; A adaptive mutation operator is proposed in the mutation operation to avoid the immature convergence; A new method is designed in the forming of the new population. The result of experiment shows that the new algorithm can greatly improve the speed and get the better quality than the traditional algorithm.Programs were all compiled in the Windows XP by MATLAB 6.5.