Obtaining proper initial clusters for the fuzzy c-mean algorithm for color reduction
This paper introduces a novel initialization method for fuzzy c-mean (FCM) algorithm for clustering color images. FCM algorithm can give suboptimal solution, if the initial centers are almost optimal. However, the problem is to find the appropriate initial centers, which are commonly, chosen by random. We apply principal component analysis (PCA) to estimate the initial centers. PCA can find the axes along the color points, which are spread in color space. The first PCA vector shows the axes with the highest distribution of its color points so it can be used as the direction to find initial centers. At first, the 3-dimensional color points should be mapped on the first PCA. Then to obtain the dominant colors, the probability density function (pdf) of the new data is calculated, the points with the highest pdf values have more chance to be initial centers. By selecting a center, the data closed to it in a diameter of σ are eliminated. The term σ is estimated based on the number of clusters. The experimental results show that the proposed method performs well for clustering color images.
color clustering fuzzy c-mean clustering initial centers
S. Gorji Kandi M. Amani Tehran
Textile Engineering Department, Amirkabir University of Technology, Tehran, Iran
国际会议
杭州
英文
110-113
2007-07-12（万方平台首次上网日期，不代表论文的发表时间）