Placing User-Generated Photo Metadata on a Map

Abstract

Abstract-In this paper we analyze large user photo collections from Flickr in order to select the most appropriate tags to describe a geographical area. We cluster photos based on their latitude and longitude and divide large areas into smaller clusters, which we will refer to as "geo-clusters". Geo-clusters have a fixed size and are able to overlap. They do not cover the entire area of interest, omitting parts where no single photo has been geo-tagged at. Within each geo-cluster we analyze all collected textual metadata i.e. the user selected tags of the photos it contains. We are then able to rank them and select the most appropriate that are able to describe landmarks and other places of interest that are contained within. Finally we place these tags on a map to help users to intuitevely understand places of interest/visual content at a glance.

A Survey of Geo-tagged Multimedia Content Analysis within Flickr

“…They used a database of more than 2M photos taken from 40 cities. In previous work [62] we suggested a probabilistic framework which aimed to place Flickr data on a map based on their tags.…”

From Tags to Trends: A First Glance at Social Media Content Dynamics

“…More recently, Kalantidis et al [17] proposed a visual-based photo image retrieval and localization approach, which exploited low-level image characteristics similarities in order to achieve accurate results. In our own most recent work on the topic [19], we proposed a tag-based methodology analyzing large Flickr user photo collections in order to select the most appropriate set of tags to be descriptive enough so as to describe an entire spatial area. Another interesting approach is [16], where meaningful travel route recommendations are proposed utilizing Flickr's user histories and past actions behaviors.…”