Image Segmentation Based On The Normalization Cut With The Clustering Algorithm

Image segmentation has been widely applied in image analysis for various areas such as biomedical imaging,intelligent transportation systems and satellite imaging.Image segmentation is to extract the meaning of the image,which is the basis of the image recognition,analysis and understanding.At present,the research of image segmentation based on graph theory is still in the primary stage,among them,the application of normalized cut method is very extensive.In this dissertation,the main algorithm is also based on the normalized cuts method.The main contributions of this dissertation are summarized as follows:Firstly,due to the normalized cut requires massive matrix of similarity measurement for image segmentation,and the calculation of large matrix is not realistic.In order to obtain high resolution information from images,the image is pre processed in this paper,the image is divided into equal sized cells,and the image is processed separately in each cell.Thus,not only loss of image information and can improve the speed of image segmentation.Then,using the K-means algorithm for image pre-segmentation.Because the algorithm is relatively sensitive to the initial clustering center,different choices often tend to result in different clustering results.Therefore,this dissertation has made the improvement for the problem.Finally,the improved K-means algorithm is nested in normalized cut algorithm.The combination of algorithms is used to segment image,it can effectively avoids insufficiency caused by using an algorithm separately and the final segmentation result is ideal.Experiments are carried out using the images of Berkeley university database and compared with other methods of image segmentation based on graph theory.By comparing the experimental results,it can be found that the algorithm can reduce the running time,and the segmentation results are satisfactory.Experimental results show that the proposed algorithm is effective and feasible.