Special session "Identifying and Linking Interesting Content in Large Audiovisual Repositories” at ACM International Conference on Multimedia Retrieval (ICMR’2017)
6-9 June 2017
As technologies for component feature identification and standard ad hoc search mature, a key challenge of multimedia retrieval is to develop mechanisms for richer content analysis and representations and modes of exploration. For example, to enable users to serendipitously create their own personal narratives by seamlessly exploring (multiple) large audiovisual repositories at the segment level, either by following established trails or creating new ones on the fly. The concept of creating networks of linked video segments within video archives that can be traversed is closely related to what some call the emerging “Web of Video” or the “Visual Web”.
Given the sheer quantities of data now becoming available in audiovisual (AV) repositories and the indefinite number of possible segments therein, one of the main challenges for multimedia exploration is to identify significant elements within this data: which AV segments are interesting enough as material for further use, for example to serve as nodes in a network of linked videos. A key research question in this area is thus whether we can automatically identify video segments that viewers would perceive to be interesting taking multiple modalities into account (visual, audio, text). Visual salience, speech, social media streams or combinations of these can be regarded as a source to derive potential video interestingness. The active research interest in this topic, in the context of movies, is demonstrated by the popularity of the “Predicting Media Interestingness” task, and earlier in the context of user-generated videos, in the “Anchoring” task, both in the MediaEval evaluation benchmark campaign.
Having identified the significant elements (e.g. anchors or hotspots) in the data the next step is to target enhanced exploration modes. The “Video Hyperlinking” task, held at TRECVid and MediaEval, focus explicitly on the creation of links between an interesting video segment and relevant target video segments. Relevance is here based on a topical relationship, using information from both the audio and visual channels.
In a special session at ICMR2017 on Identifying and Linking Interesting Content in Large Audiovisual Repositories, we invite contributions that focus on the exploration of large multimedia archives via automatically generated pathways, especially on the topics of predicting interestingness and video hyperlinking. We also strongly encourage submissions covering other relevant perspectives in this area including:
(c) GdR 720 ISIS - CNRS - 2011-2015.