| Image is one of the main ways for people to obtain information from the outside world.Digital image processing is becoming an open active research area under the increasing development of computers.Image segmentation plays a very important role in the field of image processing.The quality of image segmentation will directly affect the accuracy of subsequent research.An image can be transformed into a graph in graph theory.Therefore,based on various excellent theories in graph theory,the proposed image segmentation method has attracted the attention of scholars.The graph cut algorithm is a representative segmentation method despite of one disadvantage which is high complexity.This paper proposes a new super-pixel algorithm combining it with the graph cut allgorithm to perform image segmentation.The main work of the thesis includes the following aspects:Firstly,it introduces the basic knowledge of clustering and graph cut algorithms,including K-means clustering algorithm,fuzzy C-means clustering algorithm(FCM algorithm),Density-Based Spatial Clustering of Applications with Noise(DBSCAN algorithm)and Density peak clustering algorithm(DPC algorithm)and graph cut(GC)algorithm based on graph theory.Secondly,based on the advantage of DPC algorithm which is complete clustering quickly and accurately without iteration,the computation method of density and distance in the algorithm is improved,and Density peak clustering algorithm for super-pixel segmentation(DPCS)is proposed.The experimental comparisons with three representative super-pixel algorithms indicate that the DPCS super-pixel algorithm reduces the segmentation time and ensures the superposition of the boundary of the super-pixels,so that it successfully builds a foundation for the subsequent image segmentation processing.Thirdly,a new image segmentation algorithm by combining DPCS algorithm and GrabCut algorithm(DPCSGC)is proposed.This algorithm uses the DPCS algorithm to pre-process the image by super-pixels and then replaces a large number of pixels in the original image with super pixel blocks,therefore it greatly improves the computation speed of GMM parameter estimation and reduces the processing time of image segmentation.Fourthly,the DPCSGC algorithm and other five related image segmentation algorithms are applied to compare the results in the Berkeley image library.The result shows that the DPCSGC algorithm ensures the accuracy of image segmentation and effectively improves the segmentation speed.Finally,it is applied to the segmentation of aerial images of power system insulators.The fast and accurate segmentation result shows that the new DPCSGC image segmentation algorithm has certain value of practical application. |
