Abstract—This paper presents a new approach for image segmentation by applying Pillar-Kmeans algorithm. This segmentation process includes a new mechanism for clustering the elements of high-resolution images in order to improve precision and reduce computation time. The system applies K-means clustering to the image segmentation after optimized by Pillar Algorithm. The Pillar algorithm considers the pillars’ placement which should be located as far as possible from each other to withstand against the pressure distribution of a roof, as identical to the number of centroids amongst the data distribution. This algorithm is able to optimize the K-means clustering for image segmentation in aspects of precision and computation time. It designates the initial centroids’ positions by calculating the accumulated distance metric between each data point and all previous centroids, and then selects data points which have the maximum distance as new initial centroids. This algorithm distributes all initial centroids according to the maximum accumulated distance metric. This paper evaluates the proposed approach for image segmentation by comparing with K-means and Gaussian Mixture Model algorithm and involving RGB, HSV, HSL and CIELAB color spaces. The experimental results clarify the effectiveness of our approach to improve the segmentation quality in aspects of precision and computational time. Keywords—Image segmentation, K-means clustering, Pillar algorithm, color spaces. I. INTRODUCTION HE image segmentation is an effort to classify similar colors of image in the same group. It clusters colors into several groups based on the closeness of color intensities inside an image. The objective of the image segmentation is to extract the dominant colors. The image segmentation is very important to simplify an information extraction from images, such as color, texture, shape, and structure. The applications of image segmentation are diversely in many fields such as image compression, image retrieval, object detection, image enhancement, and medical image processing. Several approaches have been already introduced for image segmentation. The most popular method for image segmentation is K-means algorithm [1][2][12]. It is widely a A.R. Barakbah is a lecturer in Electronic Engineering Polytechnic Institute of Technology, Surabaya, Indonesia. Currently, he is a doctoral student at Graduate School of Media and Governance, Keio University, Japan (corresponding author to provide phone: +81-80-3516-4979; fax: +81-466- 486-093; e-mail: ridho@mdbl.sfc.keio.ac.jp). Y. Kiyoki is a professor of the Faculty of Environmental Information, Keio University, Japan. He is a supervisor of Multi Database and Multimedia Database Laboratory, Keio University, Shonan Fujisawa Campus, Japan (e- mail: kiyoki@sfc.keio.ac.jp). used algorithm for image segmentation because of its ability to cluster huge data points very quickly. Hierarchical clustering is also widely applied for image segmentation [20][21][24]. Many researches used Gaussian Mixture Model with its variant Expectation Maximization [9][15]. This paper proposes a new approach for image segmentation that utilizes Pillar Algorithm to optimize K- means clustering. The Pillar algorithm performs the pillars’ placement which should be located as far as possible from each other to withstand against the pressure distribution of a roof, as identical to the number of centroids amongst the data distribution. It designates the initial centroids’ positions by calculating the accumulated distance metric between each data point and all previous centroids, and then selects data points which have the maximum distance as new initial centroids. The segmentation process by our approach includes a new mechanism for clustering the elements of high-resolution images in order to improve precision and reduce computation time. In this paper, Section 2 describes the K-means algorithm. Our approach will be discussed in Section 3. Section 4 describes the experimental results using several color spaces with two comparing algorithms, and then followed by concluding remarks in Section 5. II. THE BASIC THEORY OF K-MEANS CLUSTERING This section briefly explains the basic theory of K-means clustering. Let A={a i | i=1,…,f} be attributes of f-dimensional vectors and X={x i | i=1,…,N} be each data of A. The K-means clustering separates X into k partitions called clusters S={s i | i=1,…,k} where M ∈ X is M i ={m ij | j=1,…, n(s i )} as members of s i , where n(s i ) is number of members for s i . Each cluster has cluster center of C={c i | i=1,…,k}. K-means clustering algorithm can be described as follows [26]: 1. Initiate its algorithm by generating random starting points of initial centroids C. 2. Calculate the distance d between X to cluster center C. Euclidean distance is commonly used to express the distance. 3. Separate x i for i=1..N into S in which it has minimum d(x i ,C). 4. Determine the new cluster centers c i for i=1..k defined as: ∑ = ∈ = ) ( 1 1 i s n j i ij i i s m n c (1) 5. Go back to step 2 until all centroids are convergent. A New Approach for Image Segmentation using Pillar-Kmeans Algorithm Ali Ridho Barakbah and Yasushi Kiyoki T World Academy of Science, Engineering and Technology 59 2009 23