Clustering Integration Technology To Improve The Research And Application Of Affine Propagation Algorithm

Clustering algorithm has been widely applied in various fields, such as computer technology, medicine, industrial production, digital images processing, biology and so on. In this thesis, we adopt a method proposed recently, which is called affinity propagation clustering algorithm. This method makes up consensus function instead of k-means algorithm. Affinity propagation clustering algorithm can quickly cluster data set, its algorithmic complexity is low and does not need to point out the cluster center beforehand. Affinity propagation clustering algorithm is more efficient than the k-means algorithm, it can deal with the large set of clustering problem grateful. Because of its own preferences, Affinity propagation is easy to produce shock or operation iterations too much.This paper introduces the structure of the clustering ensemble algorithm and its traditional methods in detail. Clustering ensemble focused on generating clustering ensembles and designed consensus function. In general, the k-means algorithm is used to generate multiple clustering members for the traditional clustering ensemble. Research shows that k-means is so hard to choice clustering center that the cluster results are easy to fall into the local minimum. K-means generates clustering ensemble which could not adequately reflect the characteristics of the data set. This thesis chooses bagging algorithm to sample data subsets, and the k-means algorithm was used to deal with data subsets. This method could meet the requirements of the clustering ensembles, while reducing the cost of computation.In this thesis, we studies improved affinity propagation which in visual keywords image retrieval. First, we use the SIFT algorithm extracting local feature vector set, then the improved affinity propagation clusters the feature vector, and we get the visual keywords. Through experiment contrast, improved affine propagation algorithm of visual keywords image retrieval can effectively improve the accuracy of image retrieval.