Indexing Multimedia Documents

The knowledge about the dynamic properties of multimedia documents, not only enables users to query and retrieve them, but many essential functionalities, such as (1) object prefetching for interactive document visualization, (2) result summarization/visualization, and (3) query processing for document retrieval, depend on its

• efficiency in extracting and representing dynamic information,
• speed in comparing two documents using this information, and
• capability of providing a meaningful similarity value as a result of the comparison.
  

In most non-trivial models, dynamic properties are declared as relationships between the media objects. Therefore, a general model must enable us to compare two documents based on the declared intentions of the document authors.

[back]

Most current multimedia object and document management systems have major shortcomings. First of all, existing multimedia document authoring and presentation systems and standards are lacking the flexibility required for handling the fuzziness in the temporal, spatial, and interaction structures of documents. Second, many of the systems that are used for storing and retrieving media objects and documents are still using traditional query specification and processing techniques. Furthermore, many of the proposed solutions are single-media and feature based.

We see that a symbiosis between a document management system and a document- and object-base will be beneficial for both of these systems: (1) a document system is required by the document-base to process inputs and to help visualizing query results in the form of an interactive presentation that will guide users within the solution space. On the other hand, (2) a document- and object-base is required by the document management system as a provider of media content and as a storage means for cached objects and documents.

Our interest is in developing techniques for (a) multimedia document authoring and presentation, (b) indexing, querying, and retrieval of multimedia objects and documents, and (c) result visualization tasks.

 

MODB/MIS is our view of how a storage and retrieval system for multimedia documents should be constructed. MODB should provide media object/document specification, location, and structure extraction/summarization services to MIS. MODB queries can be grouped under three categories: (1) keyword/text-based queries, (2) feature-based queries, and (3) structure-based queries. MODB should retrieve multimedia documents based on their temporal, spatial, or interaction properties. Query results must be ranked and visualized uniformly. To create interactive result visualizations, MODB uses the MIS document authoring services. The flexibility provided by the MIS presentation manager will allow a given presentation template to absorb the results of a query without causing inconsistencies that lead into incoherent visual results.

The knowledge about the dynamic properties of multimedia documents, not only enables users to query and retrieve them, but many essential functionalities, such as (1) object prefetching for interactive document visualization, (2) result summarization/visualization, and (3) query processing for document retrieval, depend on its

• efficiency in extracting and representing dynamic information,
• speed in comparing two documents using this information, and
• capability of providing a meaningful similarity value as a result of the comparison.
  

In most non-trivial models, dynamic properties are declared as relationships between the media objects. Therefore, a general model must enable us to compare two documents based on the declared intentions of the document authors.

We develop appropriate techniques that will enable real time computation of document similarities and dissimilarities, indexing of documents based on these measures, extending these measures to user interactions, and processing of the queries in the presence of inexact matches. We also consider novel algorithms for document summarization and scaling for QoS management (based on document similarity concept) as well as the associated buffer management and object/document prefecthing algorithms.

[back]