The vast amount of geo-tagged social images has attracted great attention inresearch of predicting location using the plentiful content of images, such asvisual content and textual description. Most of the existing researches use thetext-based or vision-based method to predict location. There still exists aproblem: how to effectively exploit the correlation between different types ofcontent as well as their geographical distributions for location prediction. Inthis paper, we propose to predict image location by learning the latentrelation between geographical location and multiple types of image content. Inparticularly, we propose a geographical topic model GTMI (geographical topicmodel of social image) to integrate multiple types of image content as well asthe geographical distributions, In GTMI, image topic is modeled on both textvocabulary and visual feature. Each region has its own distribution over topicsand hence has its own language model and vision pattern. The location of a newimage is estimated based on the joint probability of image content andsimilarity measure on topic distribution between images. Experiment resultsdemonstrate the performance of location prediction based on GTMI.
展开▼
