Researches On Image Segmentation Approaches

Image segmentation is very essential and critical to image processing. It is the base of target recognition and image understanding. Image segmentation has been attracting wide attention for decades because of its great importance. In image segmentation problem, the typical goal is to extract continuous regions and interest objects in the case of image processing. Some techniques in segmentation of static images, which is the most common and basic techniques in image segmentation, are concerned in this paper. Two kinds of segmentation approaches are discussed, including segmentation based on Mean Shift algorithm and based on graph theory.The Mean Shift algorithm for segmentation is a statistical iterative algorithm based on kernel density estimation. Mean Shift algorithm has been widely applied for its simplicity and efficiency. But the algorithm has some deficiencies in feature combination and image processing for large data. According to the deficiencies of the Mean Shift algorithm, this paper optimizes the structure of the algorithm for segmentation. Firstly, this paper introduces a method of data compressing by merging the nearest points with similar properties into consistency regions. Secondly, Dempster-Shafer (D-S) theory is introduced to optimize the combination of features. Lastly, in comparison with other approaches, the efficiency of the improved approach for image segmentation is validated.Image segmentation based on graph theory is a main kind of segmentation methods. The segmentation based on graph cuts is deeply discussed in this paper, and a comparison is drawn among different kinds of segmentation approaches based on graph theory. Normalized Cuts is one of the most important algorithms in all algorithms based on graph theory and it translates the segmentation problem into eigenvalue solution problem. One important problem in Normalized Cuts is how to describe an image with its context information. Based on region merge application, region adjacency graph is introduced and applied to forming an image segmentation tree by adopting Normalized Cuts.Finally, the applications of the two approaches for practical systems are implemented. And the results of experiments show the performance with the two segmentation approaches.