A Study Of Image Segmentation Technique Based On Genetic Algorithm

With the rapid development of computer technology, the degree of informatization and digitization of human society becomes much higher than before. At the same time, image segmentation technology, a key technology of digital image processing technology, is also in fast development. As the images to be segmented become more and more complex, the limitation of the traditional solutions becomes increasingly prominent and the search space of combinatorial optimization problems is increasing rapidly. The traditional method is “time-consuming effort”, and it is very difficult to obtain the optimal solution.The Genetic Algorithm(GA) obtains a good ability of global optimization, and can do optimized search only need the guide of fitness function, without depending on other auxiliary information. These abilities provide us a general framework for solving complex system problems. Therefore, Genetic Algorithm(GA) is one of the best means for solving the complex problems in the field of image segmentation. This paper studied systematically the image segmentation technology based on Genetic Algorithm, and the main research results are as follows:1. The basic theories of image segmentation techniques were studied, especially the research of the current three categories of image segmentation techniques, which are threshold segmentation technology, the edge segmentation technology and the region segmentation technology. Furthermore, the theory implementation process and the actual segmentation effect were verified.2. The Genetic Algorithm(GA) was studied deeply. Through the study on mathematical theory of Genetic Algorithm, algorithmic solving process, the basic operation of the algorithm, the room for improvement and the prevention of algorithmic immature convergence etc., the theory and characteristics of GA were more profoundly comprehended and theoretical foundation was laid for follow-up studies.3. The genetic segmentation algorithm, which based on the maximum between class variance(OTSU), was researched, especially researching its theory and algorithmic realization. And according to the conclusion of the research results, a new image segmentation algorithm was proposed, which based on the combination of improved Genetic Algorithm and improved OTSU algorithm.4. Through the research on the fitness function of the original algorithm, aiming at its shortage, the new fitness function formula was put forward, which puts the ratio of image between-class variance to within-class variance as a new fitness function. This has been verified by experiments. Experiments show that the new fitness function than the original fitness function better suited to assess the merits of the individual. And the new fitness function is more suitable for guiding population selection of GA. This improvement measures to make genetic evolution populations have been better information to guide, help speed up the algorithm in the global feasible solution space optimization speed.5. The parameter system of the original Genetic Algorithm was studied, and the relationship among the parameters has been verified by experiment. According to the results of the study, a new parameter system which can self-adaptively adjust the parameters was put forward. To be specific, that is to choose different parameter values on the basis of different evolutionary stages, furthermore with the use of the parameter system, to achieve the purpose of accelerating genetic evolutionary optimization. Through the experiment, it was successful in improving the parameter system.6. Improvements were made to the original algorithm termination condition. The original algorithm termination condition is constant and isn’t appropriate for the complex and volatile situations in realistic practice. The improved algorithm can “intelligently” decide whether to terminate evolution and output results.7. Through lots of experiments, this paper verified the three improvement measures of the new algorithm, and experimentally analyzed the relationship among these three improved measures. By experimentally contrasting with the classic genetic segmentation algorithm based on OTSU, the improved new algorithm was proved to be faster and more stable on evolutionary convergence, and to realize digital image segmentation better.