Investigation Of Clustering Algorithm Based On Spatial Domain On Image Segmentation

Image segmentation can be regarded as re-dividing and classifying theimage pixels, which is a process to make every part have their own characteristicsand satisfy the original demand after it. It is the basis of further to understand imagesand has crucial importance in the process of image analysis. Thresholdsegmentation, edge detection methods, the region growing method and segmentationmethod based on the specific theory are commonly used methods of imagesegmentation. Researchers continue to improve the existing image segmentationmethods and integrate new theory and new method of some other subjects into. Eachof these methods has advantages and disadvantages, but the method which can besuitable for all image segmentation method has not been put forward.Clustering algorithms can be divided into hard clustering algorithms, fuzzyclustering and possibilistic clustering. The clustering algorithm has been widelyused in image segmentation, and continued to be improved and optimized, andachieved good results. But when the clustering algorithm is used in imagesegmentation, it doesn’t consider the spatial information of pixels, only uses the grayinformation, the segmentation model is not complete. This thesis studies the fuzzyC-means clustering algorithm and the possibilistic C-mans clustering algorithm, thespatial information is introduced into the algorithm to build a clusteringalgorithm which is based on spatial domain. It mainly includes the following twoparts:(1) Proposes a new image segmentation method which is introduced Markovrandom field into the possibilistic C-means clustering. We use the advantagesof Markov random field that can describe the relationship between each pixel andits neighborhoods’ pixels. It can solve the problem that the possibilistic C-means algorithm does not consider the spatial pixels. It ameliorates the phenomenonof over segmenting in possibilistic C-means algorithm in multi object imagesegmentation.(2) Combine particle swarm algorithm and fuzzy C-means algorithm to imagesegmentation. Use particle swarm algorithm with dynamic inertia weighting factor,and improve the velocity updating formula, the fuzzy C-means algorithm can becombined with the whole image space information in search of the clustering center.It can obtain better segmentation effect, and decrease the number ofiterations effectively. It improves the efficiency of the algorithm.