| As one of the most important and basic field in image processing and computer vision fields, image segmentation is widely applied in practice. Presently, there are several problems in the existed algorithms when applied in the practical application, such as time consuming, the vaguely, rupture, or important details are ignored. Aimed at these problems, this dissertation expands a series of researches. The main works can be organized as follows:Above all, based on the analysis and summary of the existing algorithms of image segmentation, the typical threshold segmentation algorithm, Otsu, is chosen as the center point of the research, and combined with the global intelligent optimization algorithm-the genetic algorithm, as the starting point of further research.Secondly, considering the problem that the two-dimensional Otsu adaptive threshold algorithm is time-consuming, an improved two-dimensional Otsu threshold automatic segmentation algorithm is proposed. By calculating two 1D Otsu's threshold algorithm instead of the traditional 2D Otsu's threshold algorithm, the complexity of the algorithm is reduced from O ( L4 ) to O ( L ). In order to guarantee the integrity of the object, the minimum within-cluster scattered degree is added into the proposed algorithm, and the genetic algorithm is used to realize automatic optimize the parameter. Theoretical analysis and experimental results show that this improved method is better than the traditional 2D Otsu not only in the computation time, but also in the quality.Finally, in order to overcome time-consuming of the Otsu algorithm better, this dissertation combines with an improved genetic algorithm to find the best threshold. And in the new algorithm, the aggregation degree of the colony is brought as the divided scale to adjust the cross and variation probability adaptively. The experimental results show that the new algorithm can avoid the prematurity better, possess a well real-time, and has a better segmentation results. |
PDF Full Text Request