Pain the City Colorfully: Location Visualization from Multimple Themes


Quan Fang, Jitao Sang , Changsheng Xu and Ke Lu






The prevalence of digital photo capturing devices has generated large-scale photos with geographical information, leading to interesting tasks like geographically organizing photos and location visualization. In this work, we propose to organize photos both geographically and thematically, and investigate the problem of location visualization from multiple themes. The novel visualization scheme provides a rich display landscape for location exploration from all-round views. A two-level solution is presented, where we first identify the highly photographed places (POI) and discover their distributed themes, and then aggregate the lower-level themes to generate the higher-level themes for location visualization. We have conducted experiments on a Flickr dataset and exhibited the visualization for the Singapore city. The experimental results have validated the proposed method and demonstrated the potentials of location visualization from multiple themes.






We formulate the problem of location visualization from multiple themes as follows. The input is a set of photos within the target location, associated with their capturing positions, textual metadata and annotator user IDs. Our solution for LVMT contains three components: (1) POI identification, (2) POI theme discovery (3) location theme aggregation and visualization. Since the geographical position is only available for part of the photos, we conduct POI identification by first constructing POI vocabulary from the geo-tagged photos and then estimating the belonging POI for the rest photos. POI theme discovery is the core component. We propose an incremental learning scheme for automatically discovering the underlying themes in a POI. For location theme aggregation and visualization, we extract location themes by aggregating similar POI themes via clustering based method. With the discovered POI and location theme, we can easily visualize a POI and location from multiple themes. We have conducted experiments on a Flickr dataset of the Singapore photos.




We conduct the experiments on the dataset collected from Flickr about Singapore. This dataset contains 110,846 images including 26,623 geo-tagged photos and 9,044 users.

POI Identification


We illustrate the distribution of mined POIs in Singapore. The above figure presents the distribution map of geo-tagged images, which follows the perfect contour of Singapore. Each small circle denotes a photo, while each big circle denotes a detected POI. The photo density around a POI indicates the POI popularity.

POI Themes Discovery


At POI level, the above figure shows the salient tags and exemplary images of the Top-3 POIs in Singapore detected by our approach. For each POI, we present 3 popular themes with the salient tags and exemplary images. Clearly, these themes are related to nature scene, landmark, animals, food, and culture. We can see that the extracted images and tags are consistent and the discovered themes are adequate to represent the POIs.

Location Visualization


We exhibit the visualization of Singapore from multiple themes in above figure. Seven representative themes are presented, which correspond to scene, landmark, people, airplane, culture, animals birds and flowers respectively. Each aggregated theme is denoted by its salient tags and enriched by the belonging POIs and the associated images. From this visualized location graph with POIs and multiple themes, we can have easy access to an intuitive geographical exploration and a better thematic understanding of Singapore.




Paint the City Colorfully: Location Visualization from Multiple Themes. [pdf] [slides]

Quan Fnag, Jitao Sang, Changsheng Xu and Ke Lu
In In 19th International Conference on Multimedia Modeling (MMM), Huangshan, China,, Jan. 2013, pp.92-105.