Influence of Distance Function on Clustering Results of Affinity Propagation Algorithm
Jian Cun Zhao, Lei Xiaopei, Yu Xie · International Conference on E-Business and E-Government · 2012
Classical affinity propagation algorithm mostly takes Eular distance as a measure to estimate the distance among datas which we are interested in. Distance between every two datas can be as a elements of a matrix, so N datas can make up of a N×N matrix. We use this N×N matrix to computer the cluster results according to iterative principles. For a long time, people treat every dimension difference as equal, therefore sometimes we would obtain undesirable results. This paper analyses the influences by different distance functions on the basis of classical affinity propagation algorithm. Here we choose three distance function, Euler distance, Feature distance and Manhattan distance, which we used to cluster in image grays. The result show that: if we use different distance function to computer the same image grays, we can get different cluster result, so on, if we use different distance function to computer the same data samples, we can get different cluster result.