Research And Comparison Of Several Kinds Of Clustering Algorithm For Image Segmentation

Image segmentation is a very hot research topic at present, as an important link in image processing. It is widely used in medical, transportation, military and other related fields of computer vision, so for the study of image segmentation has important practical significance. The quality of image segmentation determines the quality of the image analysis, recognition and understanding. Currently, people put forward a lot of methods for image segmentation in order to solve the problems of application. However, Image segmentation is a very complex and difficult technology. People still need to continue the research of image segmentation technology. This paper is mainly introduced the application of image segmentation based on the clustering algorithm. Introduce the principle and the algorithm process of k-means clustering algorithm, the fuzzy c-means clustering algorithm and mean shift clustering algorithm and its improved algorithm. Then analyze them through the experiments.K-means clustering algorithm was widely used because of its simple algorithm process.But it is sensitive to the initial clustering center. According to the distribution of the random function in mathematics knowledge, the simple mean and the standard deviation are two important data to evaluate the data distribution. So the k-means clustering algorithm based on average- the standard deviation uses the sample mean and standard deviation of the object to get the more reasonable location of the initial value. Through experimental analysis, the improved algorithm on the convergence speed is greatly improved, but the difference on the accuracy of segmentation is very small. Fuzzy c-means clustering algorithm successfully introduce the fuzzy concept to the process of image pixels classification, fuzzy algorithm can retain more information of original image, compared with k-means clustering algorithm can get more accurate classification. But it also has many shortcomings. First of all its classification number need to determine beforehand, the feature recognition is not accurate and so on. Fuzzy c-means clustering algorithm based on kernel function, the pending samples in the feature space is mapped to high-dimensional feature space, makes the recognition and classification of characteristics of samples more accurately. Through experimental analysis, the improved algorithm on the classification accuracy is improved, and the convergence speed is accelerated. Mean shift algorithm does not require any prior knowledge when used in image segmentation, is an efficient clustering algorithm. But the choice of bandwidth has a larger effect on algorithm performance. The mean shift algorithm based on nearest neighbor calculates the adaptive bandwidth according to different regional characteristics by the thought of the nearest neighbor. Through experimental analysis, the improved algorithm is more reasonable on image classification, but the convergence speed limit has fallen.