A Hybrid Agglomerative Method for Improved Image Segmentation

This paper proposes a hybrid method of image segmentation by using k-means and agglomerative methods of image segmentation. The K-means method is used to find optimum number of clusters with the help gap method and a validity measure. Then this value is used as a limiting value in merge algorithm. The performance of algorithm is measured using a validity index which is measured by two factors. The first factor is intra-cluster distance whose minimum value is desired and another is inter-cluster distance for which a maximum value is required. Once optimum number of cluster is found then k-means clustering algorithm is again applied to generate large number of clusters, then from these large numbers of clusters, pair of clusters with most similar characteristics are merged iteratively until number of clusters are reduced up to optimum number of clusters. The similarity measure is taken from Davies-Bouldin Index. The proposed algorithm is performing better than simple k-means algorithm.