Abstract: A new immune clonal clustering algorithm based on the clonal selection optimization of the artifical immune system is proposed for solving unsupervised classification and recognition problems. The new algorithm can carry out the global search and the local search in many directions around the same antibody simultaneously, and make the most of antibodies in population adequately in order to search for global optimal cluster centers in the feature space quickly. It avoids the local optimum of the classical clustering algorithm. Theoretical analysis shows that the new algorithm can converge to the global optimum. Experiments on seven benchmark clustering problems of artificial data sets and two texture images segmentation problems show that the average correct rate of the new algorithm is higher than that of the K-means algorithm by 20.9％, and is higher than that of a genetic algorithm based clustering method by 20.3％.

Key words: immune clonal, clustering, K-means algorithm, texture image segmentation