I am a postdoctoral researcher at TELECOM Bretagne, working on the Georama project. I defended my PhD in October 2008, with a dissertation on "Conceptual Structures in Internet Image Retrieval". This site contains information about:

• my current work
• the publications I have (co)authored to date
• videos of the technical demonstrations I work(ed) on
• a CV

During my PhD, I developed a method for automatically constituting a geographical gazetteer (named Gazetiki) using geo-referenced information from heterogeneous data sources (such as Wikipedia, Flickr, Panoramio) which combines NLP techniques and statistical processing. Each element in Gazetiki is described by: a name, a longitude - latitude couple, a type and a popularity score. I am currently working on enriching the structure of the gazetteer by adding a spatial inclusion dimension. I am also generalizing the method to around 50 languages.

The automatically constituted geographical gazetteer can be useful in a variety of applications. Gazetiki was exploited in ThemExplorer, an image search engine for place names that combines a categorical restriction of queries and a CBIR functionality. One objective of Georama is to develop an improved version of ThemExplorer available to the public. The automatic annotation of geo-referenced images of landmarks is a second interesting application that can greatly benefit from the use of a geographical gazetteer and the first obtained results are very encouraging. A rich gazetteer can constitute the central piece of a proactive and personalized e-tourism service and I intend to develop this work direction in the near future.

I am also pursuing research concerning Web image retrieval, my PhD subject. I am interested in semantically enhanced content based image retrieval and I think that the combination of textual information surrounding the pictures and of a content description is interesting when dealing with large scale and noisy image databases. Another line of work concerns the automatic selection of the best content indexer for a given set of images. Finally, I am interested in developing a scalable method for an efficient image results reranking.