Cross-domain Retrieval of Temporal Sequences based on the Hidden Markov Model Space Formulation
- Hau San WONG (Principal Investigator / Project Coordinator)Department of Computer Science
- Zhi-Qiang LIU (Co-Investigator)School of Creative Media
DescriptionThis project addresses the problem of performing cross-domain retrieval on correlated temporal sequences belonging to different domains. In contrast to the objective of conventional content-based retrieval applications for temporal sequences, in which a set of entities that are similar to the query sequence fragment is retrieved, it is important to consider the retrieval of correlated, yet non-similar, temporal sequences that belong to a different domain from the query, as in the retrieval of suitable accompanying melodies for dancing motion sequences, and the retrieval of characteristic behavioural motion sequences for virtual characters in response to a gesture sequence input by a user in 3D interactive games. As a result, in cross-domain retrieval, a correspondence relationship needs to be identified between the set of query sequences and retrieved sequences, and a suitable feature representation needs to be adopted for temporal sequences that facilitate correspondence between the different domains.In view of these requirements, the main objectives of the project are to:search for optimal feature representations for temporal sequences belonging to different domains, based on the different characteristics of the respective domains; andidentify an optimal compatibility measure function between correlated sequences of different domains.
|Effective start/end date||1/04/07 → 4/02/10|