Segmentation Of Fusion Metal Image Dempster-Shafer Theory

Image processing and analysis techniques show the interdisciplinary potential in materials science, biomedical science, automatic control, aerospace and other fields. The study of application of image processing and analysis techniques in materials science, especially the study of image segmentation, not only is the basis of image processing and analysis, was the basis for various image processing method and application. And its result provides a firm basis for materials science research and development, so it has important research foreground and widespread application value.In materials science, we can segment metal metallographic image to extract and discern different components through image processing and analysis techniques. We can observe the influence of microstructural parameters on the properties of materials, and to predict material properties according to the grain size, using the characteristics, size, distribution of microstructure and a regular functional relationship between various mechanical properties and physical properties. It is very important to measure quantitatively and to seek out various parameters. We can segment the different components of metal metallographic image to extract and apply them through image segmentation techniques, so study of metal image segmentation is a very important and basic research. Metal image segmentation as an important branch of metal image processing is developing toward high quality, high efficiency, high reliability and automation.This thesis has conducted systematic, comprehensive and in-depth research from the aspects of image segmentation theory, methods, techniques and applications. This thesis makes a deep analysis and discussion according to the characteristics of the metal image and the shortage of the technology and methods and the difficulties that they have ever met. And also makes a series of models and improved methods which achieved good results.This thesis studies on the theory of Markov random field image model and segmentation method. This thesis uses Statistical dec ision-making and parameter estimation method of combination and according to the maximum a posteriori (MAP) optimization criterion determines the objective function. Finally creates a Markov random field image segmentation model. Compared with other segmentation algorithms, we can see that the general image segmentation methods only consider image gray level information, however, based segmentation of Markov random field model not only consider the gray-scale information but also takes into account the spatial information. The metal image segmentation experiments show that Markov random field image segmentation method has better segmentation results.This thesis has conducted in-depth research and analysis of fuzzy C-means clustering image segmentation algorithm. This thesis studies the selection of initial parameters and also analyes the problems existing in the fuzzy C-means clustering image segmentation algorithm. This thesis focuses on the influence of and initial cluster centers the relevant parameters such as clustering category number, fuzzy weighted index and iterative cut-off error on image segmentation results.Fuzzy C-means clustering image segmentation algorithm is more sensitive to noise. To solve this problem, we propose a new fuzzy C-means clustering algorithm.Through analysis of the image gray information and spatial information, we take advantage of image space neighborhood relations to expand the one-dimensional histogram to the two-dimensional histogram with spatial neighborhood relations. Then we design a simple distance measurement method. We proposed to set up the new cluster objective function which includes neighborhood information, through the distance measurement between the pixel value of clustering center and the value of neighborhood pixels. Eventually we can get fuzzy C-means clustering image segmentation algorithm with neighborhood spatial information. The segmentation and analysis results of metal images and simulated images show that the fuzzy C-means clustering image segmentation algorithm with neighborhood spatial information has a strong noise immunity ability and good segmentation results.According to the problem that a lot of relevant information is wasted because of not being utilized fully in common methods of image segmentation, this thesis has a deep analysis of the D-S theory, D-S theory has the characteristics of integration of multi-source information. It can fusion between all kinds of prior information and image information, also it can fusion between two images and even more. So It can make full use of image and related information. This thesis presents a new method of image fusion segmentation based on D-S theory, Markov random field and fuzzy theory, and then analyzes and discusses the basic probability assignment method of D-S theory.Although the image information each are not identical which Markov random field segmentation and two-dimensional histogram fuzzy cluster segmentation uses, they are not full use of information and some information is not accurate and some information lose. We put the difference between Markov random field segmentation results and fuzzy clustering two-dimensional histogram segmentation results as a redundant image. Then we use D-S theory to make up for defects caused by incompleteness, inaccuracy and uncertainty of information.The results of fusion and segmentation experiments of metallographic images, CT images, MR images and simulated images and show that the method of image fusion segmentation which based on D-S theory, Markov random field and fuzzy theory can improve the accuracy of segmentation and quality. And the method has a strong anti-noise capability. It’s a automatic, stable method so it has extensive research and application value.In a word, the development of image segmentation technology will promote the research and the application of materials science, it also promotes the further development of materials science. And it get more extensive application in various fields.