The Research Of Preprocessing And Clustering Method On Image Socialized Tag

With the continual development of network technology and digital equipment technology, more and more network image and new words fill the Internet. At this time, folksonomy as a new information organization begin to cut a figure, however, the socialized tag has its own limitations, to a certain extent the defect has hindered the development of the folksonomy.One of the socialized tag’s defects is the spelling mistakes, to address this issue, an algorithm which can screen out the recommended tag to revise the spelling error tag is given, and then based on the tag co-occurrence network, we put forward a method to measure the recommended tag’s probability and then choose the maximum probability tag as the correct tag to finish the tag correction.sometimes the image’s socialized tag may not express the full semantics, we need to learn some tags from the existed tag’s co-occurrence tag to perfect the image’s semantic.In this paper, we put forward two methods of tag learning, one is use co-occurrence tag with the help of symmetric measures and asymmetric measures, the other is use the co-occurrence tag combine with Ontology.sometimes the search tag may ambiguous,the tag clustering algorithm can help us solve this problem. Our algorithm capture the semantically similar tag to expand the search tag from the existed dictionary, and retrieve the candidate image which are possibly relevant to the query.we choose the top K tags from the candidate tag according to the their relevance to the query tags, the chosen tags are clustered top-down using a graph partitioning algorithm, and the candidate images can be clustered based on the tag cluster results,at last we give the abstract of pre tag cluster and image cluster.In order to check availability of tag learning and tag clustering, we download a large scale images and then use them experiment on a finished prototype system. Experiment results indicate that the way combine tag co-occurrence with ontology is better than the way which use tag co-occurrence or ontology only, the tag cluster algorithm is better than the Flickr’s clustering application.