Discovering Context to Prune Large and Complex Search Spaces
Specifying the search space is an important step in designing multimedia annotation systems. With the large amount of data available from sensors and web services, context-aware approaches for pruning search spaces are becoming increasingly common. In these approaches, the search space is limited by the contextual information obtained from a fixed set of sources. For example, a system for tagging faces in photos might rely on a static list of candidates obtained from the photo owner’s Facebook profile. These contextual sources can get extremely large, which leads to lower accuracy in the annotation problem.
We present our novel Context Discovery Algorithm, a technique to progressively discover the most relevant search space from a dynamic set of context sources. This allows us to reap the benefits of context, while keeping the size of the search space within bounds.
As a concrete application for our approach, we present a simple photo management application, which tags faces of people in a user’s personal photos. We empirically study the role of CueNet in the face tagging application to tag photos taken at real world events, such as conferences, weddings or social gatherings. Our results show that the availability of event context, and its dynamic discovery, can produce 80% smaller search spaces with nearly 100% correct tags.
KeywordsSearch Space Discovery Algorithm Personal Photo Event Source Social Graph
- 1.Naaman, M., Yeh, R.B., Paepcke, A., Garcia-Molina, H.: Leveraging context to resolve identity in photo albums. In: Proceedings of the 5th ACM/IEEE-CS Joint Conference on Digital Libraries (2005)Google Scholar
- 3.Stone, Z., Zickler, T., Darrell, T.: Autotagging facebook: social network context improves photo annotation. In: Computer Vision and Pattern Recognition (2008)Google Scholar
- 6.Graham, A., Garcia-Molina, H., Paepcke, A., Winograd, T.: Time as essence for photo browsing through personal digital libraries. In: Proceedings of the 2nd ACM/IEEE-CS Joint Conference on Digital Libraries (2002)Google Scholar
- 7.Cao, L., Luo, J., Kautz, H., Huang, T.S.: Annotating collections of photos using hierarchical event and scene models. In: Computer Vision and Pattern Recognition (2008)Google Scholar
- 8.Gupta, A., Jain, R.M.: Managing Event Information: Modeling, Retrieval, and Applications. Morgan & Claypool Publishers, San Rafael (2011)Google Scholar
- 9.Finkel, J.R., Grenager, T., Manning, C.: Incorporating non-local information into information extraction systems by Gibbs sampling. In: The 43rd Annual Meeting on Association for Computational Linguistics (2005)Google Scholar