trieve data. Since the potential for applications like live broadcast, video-on-demand, and digital libraries is enorIn the future we envision systems that will provide video information delivery services to customers on a very large scale. mous, the challenges presented by these applications, These systems must provide customers with mechanisms to though formidable, must be met. To do so we must consider select programs of their choice from live broadcasts. Customers and understand the following issues:should also be provided with easy means of browsing and ac-• Why traditional database retrieval techniques are incessing prerecorded digital data (e.g., distributed digital multimedia libraries), and downloading data from other information sufficient for multimedia retrieval (e.g., why can we not sources. To be viable for such large information sets, these just index and retrieve images using keywords?).systems must understand customer preferences and tailor the • How spatial visual data types differ from alphanumeric available information to the customer's needs. To support this types in terms of data modeling. vision, a number of issues must be addressed and obstacles• What the roles of the user and the system are in perovercome. Intuitive interfaces, powerful query formulation and forming queries. evaluation techniques, comprehensive data models, and flexible presentation functionalities must be developed. To realize theseVisual data are perceived differently by different people. components, an effective query evaluation engine with the capa-Because of the visual nature of video data, we end up with bilities of query resolution in different content-specific formats numerous interpretations of the same data. To represent (e.g., by graphics, by image, by sound) and in different domain-all the different interpretations by keywords (text) is an specific models (e.g., database of movies, database of newsclips) impossible task as one cannot foresee all possible interpreshould be present. Additionally, the digital video database will tations of the data during indexing. Also, representation require an efficient indexing system for easy access to the stored of a small segment of video data by a large number of information. In this paper we discuss existing research trends keywords will lead to space explosion during indexing.in this area and requirements for future data delivery systems. Keywords cannot successfully represent the temporal naAn overview of video indexing is presented followed by a discusture of video data nor do they support semantic relationsion on current indexing techniques.

No abstract

Video production involves the process of capturing, editing, and composing video segments for delivery to a consumer. A composition must yield a coherent presentation of an event or narrative. This process can be automated if appropriate domainspecific metadata are associated with video segments and composition techniques are established. Automation leads to the support of dynamic composition and customization for applications such as news on demand. In this paper, we present techniques to achieve dynamic, real-time, and cohesive video composition and customization. We also identify metrics for evaluating our techniques with respect to existing manually produced video-based news. The results of such an evaluation show that the quality of automatic composition is comparable to¦and in some cases, better than¦broadcast news video composition. The results also validate the assertions on which the automatic composition techniques are based.

In this paper we describe the Virtual Video Browser (VVB) software application designed to allow the interactive browsing and content-based query of a video database and to facilitate the subsequent playout of selected titles. The VVB is a manifestation of our mechanisms for the location, identification, and delivery of digital audio and video in a distributed system which can be extended to several application domains including multimedia-based home entertainment, catalog shopping, and distance learning. The VVB employs a two phase retrieval process to serve its users. In the query phase, user queries are sent to a metadata server for processing. In the subsequent playout phase, a connection is established between the client workstation and a video server for the delivery of video data. The VVB incorporates a simple query interface that lets users specify their preferences to the system and retrieve the appropriate video. The application is designed to work in a distributed environment where video sequences are stored in different databases interconnected via a network. It has been shown to be a viable target application useful for investigating research problems related to building interactive multimedia systems.

Segmentation, video data modeling, and annotation are indispensable operations necessary for creating and populating a video database. To support such video databases, annotation data can be collected as metadata for the database and subsequently used for indexing and query evaluation. In this paper we describe the design and development of a video annotation engine, called Vane, intended to solve this problem as a domain-independent video annotation application. Using the Vane tool, the annotation of raw video data is achieved through metadata collection. This process, which is performed semi-automatically, produces tailored SGML documents whose purpose is to describe information about the video content. These documents constitute the metadatabase component of the video database. The video data model which has been developed for the metadata, is as open as possible for multiple domainspecific applications. The tool is currently in use to annotate a video archive comprised of educational and news video content.