Enhanced K-Means clustering algorithm for color image segmentation

Abstract

K-Means Clustering Algorithm is one of the most popular and unsupervised learning algorithm , this research work details the implementation of new adaptive technique for color image segmentation that is the enhanced version of K-Means Clustering Algorithm. The standard K-Means algorithm produces accurate segmentation results only when applied to images defined by homogenous regions with respect to texture and color, and the K supplied to the standard algorithm is user defined which can lead to the empty clusters in the segmentation. In addition, the initialization of the K-Means algorithm is problematic and usually the initial cluster centers are randomly picked, hence there’s a need to enhance the K-Means clustering algorithm by first predefining the optimal number of clusters and merging the similar looking cluster until it reaches to the optimal number of cluster. The main contribution of this work is the enhancement of the K-Means Clustering algorithm that includes the primary features that describe the color smoothness and the quality of the clusters in the process of pixel assignment. The resulting color segmentation scheme has been applied to a large number of natural images and the experimental data indicates the robustness of the new developed segmentation algorithm.