Abstract
In this paper, we proposed a new clustering method where each cluster is created based on its characteristic that we call texture. Extraction of the texture relies on measuring the similarity of neighboring patterns. Our proposed clustering algorithm consists of two stages. In the first stage, sub-clusters are created based on the similarity of their structures and in the second stage, the sub-clusters with similar textures and closeness of their distance are hierarchically combined to create a larger cluster. A theoretical justification for the proposed cluster isolation measure has been represented. Experimental results with complex data reveal that the performance of our method is superior than the well known K-means and Single Link methods. The proposed clustering algorithm is independent of the order of training samples appearance and its computational complexity is less than that of the traditional hierarchical algorithms.