Image Segmentation Based On Multi-agent

Image segmentation refers to the image is divided into regions with the different characteristics and extract the target of interest in technology and process. It is a key step from the image processing into the image analysis, and the basis of further image understanding. Image segmentation technology is closely related to many other subjects. With the various disciplines of the new theories and methods proposed, many segmentation technology combining the relevant theories, methods and tools appears. Image segmentation based on Multi-Agent is one of them.In the paper that adaptive image segmentation with distributed behavior-based Agents proposed by Jiming Liu, author designs Agents with the autonomous behavior for image segmentation problem. Through the breeding, diffusion of Agents gradually labels homogeneous regions, the eventual realization of image segmentation. It has achieved good results, but there still exist some problems, so this dissertation has made improvements in the following areas:1) For each pixel, its feature is the vector combined gray and gray co-occurrence matrix statistics component. It makes algorithm more applicable to texture image and S AR (Synthetic Aperture Radar) image.2) Initial Agents are classified by its feature similarity. Agents with more similar get together. The algorithm convergence rate is raised than the original initial random distribution.3) The standard of homogeneous region is defined by the adaptive evolution of agents. The algorithm is more applicable, robust and practical than the pre-defined standards in the original paper.4) A competitive operator is proposed. The competitive operator works when two kinds of agents label a pixel at the same time.Experimental results show better performance.There is a problem in image segmentation. That is contradiction that regional consistency and accuracy of the edge of the division. In this dissertation, because of the thinking of agents and the lack of information in spatial domain, an edge detection and amendment method based on transform domain is proposed. The algorithm detects edges using transform domain-based feature firstly, and then applies edge amendment operator. The operator acting on the edge, it takes the characteristics of edge's shape as a priori knowledge. Pixels near the edge are segmented according to that. The test results show the effectiveness of the method.