A Study Of Interactive Graph Cut Algorithm On Color Image Segmentation

Image segmentation is a basic and important research content of image processing. This kind of technology divides image into some characteristic areas. After that, users can extract the region of interest to do the further analysis and processing. It is applied in a variety of fields, especially in artificial intelligence, pattern recognition, computer vision, etc.Traditional manual segmentation methods are exhausting and time consuming. These methods cannot meet the increasing demand in this area. The automatic segmentation methods can greatly improve the efficiency. Unfortunately, computer artificial intelligence has not reached such a high level, which is difficult to obtain satisfactory results. Thus, interactive image segmentation methods which combine the advantages of these two methods have obtained increasing attention in recent years. These methods adopt the user’s subjective understanding as prior knowledge to guide segmentation processing. Compared with the traditional methods, the advantage of these methods is segmenting image more accurately and efficiently. The background of interactive image segmentation has been introduced in this paper. The principles and application range of the image segmentation based on graph theory have been researched. Based on this, the research content in this paper is as follows:1. Based on graph cut energy function, another definition of boundary penalty term has been proposed. It uses the local coefficient of variation of pixel to introduce the neighborhood information of the image. And then this method is applied to segment color images. This allows the label of a pixel to be influenced by not only the color information and spatial position of this pixel but also the immediate neighborhood information. Then this energy function is minimized by using maxflow/mincut algorithm proposed by Boykov and Kolmogorov. Namely, finding the minimum cut of the graph which is corresponding to the image. Finally, get the segment result according to this cut. Thus the object edge is more clearly.2. Based on graph cut energy function, multi-objective optimization method is adopted to segment the color images. And the two objective functions are the regional penaltyand boundary penalty of graph cut energy function. The Non-dominated Sorting Genetic Algorithm II, shorted for NSGA-II is used in this paper, which is the existing best method to solve multi-objective optimization problem. The tradeoff parameter λ in the graph cut energy function determines the degree of the influence of the boundary term and regional term on segmentation results. For each image, this parameter is determined by trial and error. While the method proposed in this paper avoids using any parameter.