Based On Random Walk Theory Of Image Segmentation Method

Image segmentation refers to dividing the image into different regions,each of which has a specific meaning,these regions have to meet specific regional consistency and don't overlap each other. Image segmentation in image engineering occupies very important position, which is the basis of image recognition. The image segmentation results have a great influence to the follow-up analysis and processing.Manual intervention image segmentation based on graph theory in recent years with their own advantages attracted widespread attention in research, such as the image segmentation based on random walk algorithm is a semi-automatic image segmentation method. Random walk is the first one of the random process, not only in mathematics, game, network, financial and many other areas, but also in the analysis of queuing, field theory and the areas of bankruptcy system of theory have been widely used.Random walk has a good mathematical foundation which not only has strong resistance to noise in the image segmentation, but also can detect weak edges well, so it is an effective image segmentation method, and in recent years it has developed into having most achievements in scientific research and the most abundant results in the area of combination graph theory.Random walk theory as the foundation, this paper conducted study on image segmentation, mainly in the following work:(1)This paper summarized the traditional and modern methods of image segmentation, compared their respective advantages and disadvantages, and introduced the corresponding improvement methods. The random walk method was outlined mainly.(2)This paper introduced several methods based on graph theory, such as random walk, maximum flow-minimum cut, minimum spanning tree, and given the basic content related to the random walk theory, including random walk in the straight line and plane, and the framework for image segmentation based on random walk theory, and transform solving the problem of the harmonic function into solving problems of the sparse linear equation system through the internal links of random walk and circuit theory.(3)As the speckle noise intensity, low contrast, weak borders in medical ultrasound image, the traditional methods are difficult to obtain good segmentation results. This paper described the advantages of random walk in detail. The random walk algorithm was used in medical ultrasound image segmentation, not only effectively suppress the speckle noise, but also accurately detect the target in medical ultrasound image of the weak edge, get more accurate segmentation results. And this method solves the sparse, symmetric, positive definite system of linear equations to obtain solutions of Dirichlet problem, so the calculation speed greatly improve.(4)As the traditional random walk algorithm only consider the similarity between adjacent pixels in constructing weighted undirected graph, without taking into account the surrounding pixels gradient information,thus inhibiting the random walker walking along some edges of the gray level close to the seed point to it,leading to.misclassification or omission points.And due to the traditional random walk algorithm use 4-pixel neighborhood,that is the current pixel only choose four directions of its neighbors,so the segmentation results of image is not accurate enough. This article proposed a more robust improved algorithm,which integrate gray information with gradient orientation information in image segmentation. Based on original random walk algorithm, this algorithm exploited gradient information and gray information of 8-pixel neighborhood of image together to map the edge weights of network assignment, and then marked each pixel tags to implement image segmentation.