Research Of Threshold Image Segmentation Based On Graph Theory

As one of the most important and typical problems in image processing and computer vision fields, image segmentation is the basic premise in image vision analysis and pattern recognition.Image segmentation based on graph theory is a research focus in image segmentation fields,this approaches are the formation of a weighted graph,where each vertex corresponds to an image pixel,the best segmentation of the image can be obtained by minimal cut sets,and achieve good results of image segmentation.But some problems and deficiencies may be found in the process of implementation, if we just simply use this approaches. other theories are combined with this approaches to get better effectiveness.Aiming at the problem exists in the image segmentations based on graph theory, the thesis combines the Normalized Cut standard and Min-Max Cut standard with threshold image segmentation, and improves the computational formula which used to measure the similarity between the pixels to reach a well effectiveness. The main works can be organized as follows:Part one: To make sure the feasibility of the combination of the two methods, the theories and progress of image segmentation based on graph theory and threshold segmentation were researched.Part two: Aiming at the problem in the process of the image segmentations based on graph theory, a new algorithm is presented to resist noise in the process of image segmentation. Normalized Cut standard is adopted as the standard of the division between the objective and background, and the relevant space information between pixels and its neighborhood are used to improve the noise immunity. The experimental results show that the method has well noise immunity.Part three: Aiming at the inadequate of commonality in the image segmentations based on graph theory, a new method based on Min-Max Cut threshold image segmentation is proposed. This method adopts Min - Max dividing measure as the threshold segmentation rule to distinguish the target and background. In describing the weight matrix of similarity between the pixel values of each image,we use weight matrix based on grey level to replace that based on image pixels so that the algorithm complexity and the storage space degrees are greatly saved. By using the computational formula of the potential function as the graph weight matrices, the formula can reflect the similarity between pixels better in statistics. Meanwhile, it avoids the defect by the manual forms to set the control of some similarities between the degree of sensitivity parameters, and improves the commonality of the algorithm. The experimental results show that the algorithm has been better practical.