AUTOMATIC INITIALIZATION of IMAGE CLUSTERING ALGORITHMS

Authors : Merve ARSLAN, Recep DEMİRCİ

Pages : 297-316

View : 36 | Download : 14

Publication Date : 2022-12-31

Article Type : Research Paper

Abstract :Clustering is partition of a data set into subsets where each item in assigned subset is similar and different from that of other subsets. K-means and fuzzy c-means insert ignore into journalissuearticles values(FCM); algorithms are frequently used for clustering of color image. On the other hand, randomly determination of initial cluster centers is one of the most important problems of both algorithms since results to be obtained vary according to initial values of cluster centers. Thus, obtaining different results at each run time reduces reliability of algorithms. One of a typical solution is that number of iterations is increased in order to obtain an accurate result. However, it increases computation cost. A novel solution for initial cluster centers has been proposed in this study where octree algorithm was used. Color images were initially quantized in certain numbers of color vectors depending on level of octree algorithm. Then, means of each quantized color vector set were obtained. The pixel numbers of each pre-subset were sorted and assigned as initial cluster centers. Consequently, cluster centers are selected automatically. As positions of quantized vectors in color space are fixed, a deterministic algorithm has been attained.
Keywords : K means, Fuzzy c means, Octree, Color Quantization

ORIGINAL ARTICLE URL