Research On Image Segmentation Based On Intelligent Algorithms

As the foundation of the image processing and analysis, image segmentation is an important computer vision technology. Because of its intuitive properties and simplicity of implementation, image thresholding enjoys a central position in applications of image segmentation. Fuzzy entropy thresholding is the most popular method of the thresholding segmentation.Traditional fuzzy entropy method is very sensitive to the noise. For this problem, an improved fuzzy entropy method based on the two-dimensional histogram is proposed. The improved method partitions an effective region in the two-dimensional histogram, and redefines the mebership funcction according to the relationship between the pixels and its neiborhod.Grey relational analysis in the grey system theory is introduced to the fuzzy entropy thresholding method. It is used to express the relational degree of the current pixel and its neighborhood pixels, and is associated with the membership function so that a new membership function is redefined. For the grey relational degree is computed by two methods, two different modified fuzzy entropy thresholding methods are presented.One threshold often cann't satisfies the requirment of some applications, so multi-threshold segmentation methods are necessary in some situations. However, the traditional exhaustive method has a large computational cost, which makes it difficult to be applied in the real engineering situation. Therefore, the multi-threshold searching method based on intelligent algorithms has good prospects in future.In allusion to the not strong global searching ability of the particle swarm optimization, a modified particle swarm optimization is proposed based on the random parameters strategy. In order to overcome the defects of the traditional fuzzy clustering method, the fuzzy clustering validity index function is introduced. And then, a variable-length particle swarm optimization is designed in order to realize the adaptive segmentation based on the fuzzy clustering validity index function.By adding the gene jumping operation in the standard genetic algorithm, is effective to improve the convergence speed and the global searching ability. However, the gene jumping operation of the standard jumping gene genetic algorithm is lack of guidance, which destroyes the good genes easily. Therefore, three improvement measures are investigated. And then, the jumping gene genetic algorithm is applied in the fuzzy entropy thresholding method, to fastly obtain the optimal segmentation tresholds.Ant colony algorithm is characterized by positive feedback and robustness. Ant colony algorithm is used in the miximization of the fuzzy entropy function, to obtain the optimal thresholds.A large number of testing functions and real images are used for verifying the proposed image segmentation methods based on various intelligent algorithms, the results of which show that the proposed methods provide new solving methods for the image segmentation.