Image Segmentation Method Based On Swarm Intelligence Research

Image segmentation not only is the base of image understanding, but also is a key in image analysis. Some data indicate that when more prior knowledge is used in the procedure of image segmentation, the segmentation can acquire good precision and robustness, while the time-complexity decreases. Aiming at the problem of slow speed and noise-sensibility in image segmentation, this thesis studies the technology and methods on image thresholding segmentation.In the thesis, grey theory, wavelet transform, fuzzy theory, pattern recognition, entropy, and histogram are integrated to design some new image thresholding segmentation models for high quality segmentation. At the same time, community intelligence is introduced to speed up the segmenting procedure, which is effective in parallel optimization.The main contribution can be summarized as follows:(1) A fast SAR image segmentation method based on 2D grey entropy model is suggested, which combines Genetic Algorithm, 2D entropy and grey system theory. Some experimental results indicate that the method not only ignores the disturbance of inherent speckle in SAR image, but also shortens the segmenting time obviously.(2) A fast SAR image segmentation method based on grey theory and Tsallis entropy is proposed, which employs Tsallis entropy and PSO to image segmentation. This new method not only shortens the segmenting time obviously, but also ignores the disturbance of inherent speckle in SAR image and illustrates some flexibility in segmenting different objects.(3) A fast SAR image segmentation method based on grey fuzzy entropy is presented, in which we introduce grey relational analysis and fuzzy theory to decrease the noise-sensibility of the traditional fuzzy entropy methods. In addition, PSO, as an swarm intelligent tool, is used to speed up the segmenting procedure.(4) A fast image segmentation method based on Fisher criterion and grey post processing is discussed, which utilized the Fisher criterion in pattern recognition. Some experimental results indicate that the method not only ignores the disturbance of inherent speckle in SAR image, but also shortens the segmenting time obviously.