From content-based multimedia analysis to large-scale semantic browsing.

Guillaume Gravier - Research fellow at the Centre National pour la Recherche Scientifique (CNRS), IRISA, Université de Rennes http://www.irisa.fr/metiss/members/gravier/
Multimedia Communications

Date: -
Location: Eurecom

The first part of the talk will review activities at IRISA in multimedia analysis. After a short overview of multimedia activities at IRISA, I will present a synthesis of the research we have undertaken over the past 10 years. In particular, I will mention two research tracks in the field of video analysis : joint multimodal modeling with statistical approaches and spoken content processing combining speech recognition and natural language processing. I will conclude with exploratory work on unsupervised discovery of repeating motifs in video and audio contents. The second part of the talk, from a prospective standpoint, will elaborate on the concept of large-scale semantic multimedia browsing. This concept, which consists in creating on a large scale links between multimedia documents according to their meaning, is presented as an extension of today's vision of multimedia, largely dominated by the information retrieval paradigm. We will analyze the challenges faced by large-scale semantic browsing, such as robustness and effectiveness, and propose several research directions along the three following lines: multimodality, semantic and data mining.