Modélisation de l'extraction des descripteurs visuels - Intégration de relations topologiques

Despite its major role in the automatic annotation, the visual descriptors extraction process is not explicitly modeled and the contribution of each of its steps on the quality of automatic annotation is not sufficiently studied. In this article, we first propose a model for visual descriptors extraction. Using this model, we propose the inclusion of relationships by applying a topological grouping technique of regions of interest; leading to the definition of richer descriptors. Experiments on VOC2009 corpus show that our approach significantly improves the automatic annotation results.