MedIC/CISMeF at ImageCLEF 2006: Image Annotation and Retrieval Tasks

In the 2006 ImageCLEF cross-language image retrieval track, the MedIC/CISMeF group participated at the two medical-related tasks: the automatic annotation task and the multilingual image retrieval task. For the first task we submitted four runs based on supervised classification of combined texture and statistical image representations, the best result being the fourth rank at only 1% of the winner. The architecture proposed for the second task is reposing on textual-retrieval using terms derived from the MeSH thesaurus, combined with ranking by visual similarity. Due to technical and practical difficulties, the only run we were able to submit was incomplete, resulting in a modest result in the pools. Therefore, the actual capacity of the proposed retrieval architecture could not be evaluated.