Web Image Semantic Annotation And Aggregation Based On Social Network

In the recent years, the social network based on the sharing ideas attracts the special attention of the most Internet users. As the social network of Flickr and Facebook, which provide digital sharing images, have become the image repository for most Internet users. According to the statistics, Flickr hosts more than 3 billion images with around several million new uploaded images per day. Flickr allows users to assign tags for images and recommend their images to the relevant community. So it brings a great convenience for the image management and retrieval. The social tagging makes up for the deficiency of the traditional added annotation by manual. Also it supplies a new research method for the image semantic automatic tagging.The main work of this paper is as follows:(1) We implement the image retrieval algorithm based on SIFT image fingerprint so that provides content-based image retrieval for the tag learning in the vast Web image database. Firstly, the algorithm exacts SIFT features from images and saves them to image data table. Then, in terms of SIFT features, the target image retrieves the neighbor image dataset which the visual content of image is similar to the target image. The neighbor image dataset supplies the data basis for the learning tag relevance described in next.(2) According to the characteristics of the social relationship network data, we implement the social relationship partition based on Girvan Newman algorithm. So the social media database can be divided into several media theme communities by the Girvan Newman algorithm.(3) As the uncontrolled nature of social tagging and the diversity of knowledge and cultural background of its user that the average quality of the image's existing social tags is lower. That these tags with lower quality could reduce the efficiency and accuracy of the image retrieval. So we propose an algorithm of the image tag learning based on social network. We implement the image tag relevance learning based on the nearest neighbor voting algorithm by the means of the neighbor relationship in the social network and the SIFT image retrieval algorithm.(4) Based on the study of algorithms above, we implement the Web image aggregation platform based on social network semantic annotation. The main function of the platform includes the image aggregation and retrieval module, the multimedia editor module and the personalized information customization module.With the sharing image on the social network's geometric growth, so it will become very important that how to rapidly, accurately and efficiently retrieve the available Web images from the vast of network image database. And the retrieval result can provide service to the multimedia aggregation and webpage editor module for users. Therefore, the study of the Web image aggregation platform based on social network semantic annotation has highly social value and research value.