An Optimized Clustering Algorithm for Contour Data

Abstract

The initial center’s selection of the traditional K-means algorithm is random. It makes the algorithm instability. Traditional clustering methods are hard to apply to the clustering of contour data. The data from the contour data set forms the contour shape, but the cluster center is not on the contour. The final cluster center is in the inner center of the contour. Using K-means algorithm directly is not able to handle this kind of data well. In order to optimize the initial clustering center of the contour data, and in order to solve the problem of contour data clustering. Based on the high density area may be initial clustering centers. And under the thought of the farther apart distance the more likely belonged to different clustering. In this paper, an algorithm of contour data clustering is proposed. This algorithm can obtain better initial central for contour data, so as to improve the clustering effect of contour data. Firstly, the algorithm calculates the distance between the samples. And based on the premise of the two farthest points most probability does not belong to the same class. The algorithm finds out the farthest two points according to the sample distance. With one point as the center, find the point closest to the point to join the point set. Until the data set number greater than or equal to α number (α is the ratio of the number of samples to the number of clusters). Calculate the average of all points as the initial cluster center. Repeat the above steps to get K initial cluster centers. Then the final clustering calculation is carried out according to the K-means algorithm. After the experiment on the experimental data set, the improved algorithm has a strong stability. The initial clustering center is uniformly discrete. And the clustering results have a high accuracy and have a good F1 value. The improved algorithm solves the clustering problem of contour data well.