Research On Image Clustering Based On Correlation Mining

Along with the image media devices had become almost commonplace, socialnetwork and Internet data sharing gain popularity. Now people can obtain images notonly through the camera, scanner and other acquisition equipment, but also access to alarge number of digital image information through the Internet. It is difficult to obtainaccurate information as soon as possible from the large scale image information withrich semantics. Therefore, it is essential for us to study efficient image clusteringalgorithms to achieve efficient management and use of images.How to measure the correlation among the samples is key issue in the study ofefficient image clustering algorithms, which will directly affect the results of clustering.In this paper, we propose a new image clustering method based on SIFT and affinitypropagation. Firstly, we extract image features based on Scale-invariant featuretransform(SIFT), which are invariant to image scaling, translation, and rotation, andpartially invariant to illumination changes and affine. Then the modelling of correlationis established and quantitatively analyzed. Finally, we utilize affinity propagation tofuse different SIFT correlations among images and obtain the final clustering results.Our method simultaneously considers all data points as potential exemplars.In the evaluation of the experimental results, we use clustering accuracy asstandards. The experiment results show that image clustering based on correlationmining can not only achieve good performance on various image data set, but also theclustering accuracy is higher than traditional image clustering algorithm.