Utilizing Reverse Viewshed Analysis in Image Geo-Localization

When users browse beautiful scenery photos uploaded on a social media website, they may have a passion to know about where those photos are taken so that they could view the similar sceneries when they go to the same spot. Advancement in computer vision technology enables the extraction of visual features from those images and the widespread of location-awareness devices makes image positioning possible with GPS coordinates or geo-tags (e.g., landmarks, place names). In this paper, we propose a novel method for image positioning by utilizing spatial analysis and computer vision techniques. A prototype system is implemented based on large-scale Flickr photos and a case-study of the Eiffel Tower is demonstrated. Both global and local visual features as well as the spatial context are utilized aiming at building a more accurate and efficient framework. The result illustrates that our approach can achieve a better accuracy compared with the baseline approach. To our knowledge, it is among the first researches that combine not only the visual features of photos, but also take the spatial context into consideration for the image geo-localization using high-density social media photos at the spatial scale of a landmark.