Visual Language Modeling for Mobile Localization LIG participation at RobotVision’09

Abstract

This working note presents our novel approach for scene recognition (i.e. localization of mobile robot using visual information) in the RobotVision task [1] based on lan-guage model [2]. Language model has been successfully used for information retrieval (specifically for textual retrieval). In recent study [3], this model has also showed a good performance on modeling the visual information. For this reason, it can be used to address several problems in image understanding such as: scene recognition, im-age retrieval, etc. We have developed a visual language framework to participate in RobotVision’09 task this year. This framework consists of 3 principal components: a training step, a matching step and a post-processing step. Finally, we present the re-sults of our approach on both validation set and test set released by the ImageCLEF’s organizer.