How Spatial Segmentation improves the Multimodal Geo-Tagging
Pascal Kelm, Sebastian Schmiedeke, Thomas Sikora · MediaEval · 2012
In this paper we present a hierarchical, multi-modal approach in combination with dierent granularity levels for the Placing Task at the MediaEval benchmark 2012. Our approach makes use of external resources like gazetteers to extract toponyms in the metadata and of visual and textual features to identify similar content. First, the bounderies detection recognizes the country and its dimension to speed up the estimation and to eliminate geographical ambiguity. Next, we prepared a training database to group them together into geographical regions and to build a hierarchical model. The fusion of visual and textual methods for dierent granularities is used to classify the videos’ location into possible regions. At the end the Flickr videos are tagged with the geo-information of the most similar training image within the regions that is previously ltered by the probabilistic model for each test video.