Clustering Technique Using K-means Dempster-shafer Theory of Evidence

Identification of objects from heterogeneous regions is one of the challenging tasks in image mining. It can be thought of as partitioning image into clusters based on the image attributes. The purpose of clustering is to identify the similar groupings from a large data set to produce a precise representation of the image. Clustering requires classification of pixels according to some similarity metrics. Classical clustering algorithm, K-Means finds the clusters based on similarity metric. One of the draw back with the standard K-Means algorithm is that it produces accurate results only when applied to images defined by homogeneous region and the cluster are well separated from each other by the way of randomly picking cluster center. In our approach hierarchical clustering algorithm is used to find the initial cluster center and the mass value for the pixels is determined which decides the pixel inclusion into the appropriate clusters. The experimental results show that the new clustering algorithm outperforms well.