Multimodal image geocoding: The 2013 RECOD's approach

Abstract

This work describes the approach used by the RECOD team in the MediaEval Placing Task of 2013, in which we were required to develop an automatic scheme to assign geographical locations to images. Our approach is multimodal, considering textual and visual descriptors, which are combined by a rank aggregation strategy. We estimate the location of test images based on the coordinates of top-ranked images in the list of combined results.