Novel Partitioning Clustering

Abstract

Partitioning clustering is an unsupervised approach and can segment images in a less complex fashion. Thus, partitioning clustering is used in many digital image processing fields, e.g., airborne and medical image processing. The k-means and fuzzy c-means clustering techniques are examples of popular hard and soft membership-based clustering techniques. The partitioning clustering techniques may have dead centers, trapped centroids, and outliers’ sensitivity problems. Therefore, the partitioning clustering techniques do not always converge to optimum global location. The partitioning clustering techniques have been modified to overcome the mentioned problems. First, this chapter discusses the possible enhancements in k-means clustering techniques to overcome its dead centers and trapped centroids problems and illustrates the working of two new enhanced versions of the k-means clustering techniques. The modified pixels are assigning and transferring k-means clustering techniques to confirm their final solution’s convergence at the optimum global location. Next, this chapter discusses the possible enhancements in the membership function of the fuzzy c-means technique that reduces its outlier’s sensitivity. The outlier rejection approach of advanced fuzzy c-means clustering ensures that its final solution will converge at the optimum global location.