Where your photo is taken: Geolocation prediction for social images

Social image sharing websites, such as Flickr and Zooomr, have attracted a large number of Internet users. These systems allow users to associate geolocation information to their images, which is essential for many interesting applications such as location-aware image search. However, only a small fraction of social images have geolocation information. Thus, an automated tool for suggesting geolocation is essential to help users geotag their images. In this paper, we investigate how to assist users in geotagging social images, and how to boost the accuracy of geotagging. We use a large dataset consisting of 221M Flickr images uploaded by 2.2M users. We analyze for the first time user uploading patterns, user geotagging behaviors, and the relationship between the taken-time gap and the geographical distance between two images from the same user. Our analysis shows that the takentime gaps between the image to be geotagged and historical images are very important for geotagging. Based on the finding, we represent a user profile by historical tags for the user, and build a multinomial model on the user profile for geotagging. We further propose a unified framework to suggest geolocations for images, which combines the information from both image tags and the user profile. Experimental results on the Flickr dataset show that for images uploaded by users who have never done geotagging, our method outperforms the state-of-the-art method by 10.6% to 34.2%, depending on the granularity of the prediction. For images from users who have done geotagging, a simple method is able to achieve very high accuracy.