T he rising popularity of photosharing applications on the Web has led to the generation of huge amounts of personal image collections. Browsing through image collections of such magnitude is currently supported by the use of tags. However, tags suffer from several limitations-such as polysemy, lack of uniformity, and spam-thus not presenting an adequate solution to the problem of content organization. Therefore, automated contentorganization methods are of particular importance to improve the content-consumption experience. Because it's common for users to associate their photo-captured experiences with some landmarks-for example, a tourist site or an event, such as a music concert or a gathering with friends-we can view landmarks and events as natural units of organization for large image collections. It's for this reason that automating the process of detecting such concepts in large image sets can enhance the experience of accessing massive amounts of pictorial content.In this article, we present a novel scheme for automatically detecting landmarks and events in tagged image collections. Our proposal is based on the simple yet elegant concept of image similarity graphs as a means of combining multiple notions of similarity between images in a photo collection; in our case, we use visual and tag similarity. We perform clustering on such image similarity graphs by means of community detection, 1 a process that identifies on the graph groups of nodes that are more densely connected to each other than to the rest of the network. In contrast to conventional clustering schemes such as k-means or hierarchical agglomerative clustering, community detection is computationally more efficient and doesn't require the number of clusters to be provided as input. Subsequently, we classify the resulting image clusters as landmarks or events by use of features related to the temporal, social, and tag characteristics of image clusters. In the case of landmarks, we also conduct a cluster-merging step on the basis of spatial proximity to enrich our landmark model.

No abstract

The wide adoption of photo sharing applications such as Flickr c and the massive amounts of user-generated content uploaded to them raises an information overload issue for users. An established technique to overcome such an overload is to cluster images into groups based on their similarity and then use the derived clusters to assist navigation and browsing of the collection. In this paper, we present a community detection (i.e. graph-based clustering) approach that makes use of both visual and tagging features of images in order to efficiently extract groups of related images within large image collections. Based on experiments we conducted on a dataset comprising publicly available images from Flickr c , we demonstrate the efficiency of our method, the added value of combining visual and tag features and the utility of the derived clusters for exploring an image collection.