Crossing textual and visual content in different application scenarios

In this presentation, we present a method based on Trans-media Pseudo-Relevance Feedback that allows crossing visual and textual content through multimodal knowledge base. The main idea is to use one of the modalities to retrieve multimodal documents from the knowledge base and then to switch (or to fuse with) the other modality in the next step. The different potentials of the method (retrieval, image annotation, text illustration ...) are illustrated within a Travel Blog Assistance System example scenario.