4. Multimedia Databases

The design and implementation of multimedia databases create additional challenges due to the nature of multimedia data and the requirements of possible applications. Multimedia applications can be categorized in three main groups, each of which poses different data management challenges [2]:

Repository applications: Large amounts of multimedia data and associated metadata are stored for retrieval purposes. These repositories may be distributed or centralized and can be managed using conventional DBMS. Examples of multimedia data stored in such repositories include medical and satellite images, and engineering drawings.
Presentation applications: Applications that involve delivery of multimedia content to a possibly remote location, subject to temporal constraints. In these applications, data is consumed as it is delivered, as opposed to being stored for later processing. A number of new potential problems, such as jitter, latency, and the corresponding need to maintain and guarantee "quality of service" (QoS), come into play. Examples of such applications include audio and video broadcasting over the Internet.
Collaborative work using multimedia information: a new breed of multimedia applications in which a geographically dispersed group of professionals (e.g., engineers or medical doctors) work together on a common, multimedia-intensive, task.

Accommodating massive amounts of text, graphics, images, animations, audio, and video streams into a database system is far from trivial and the popularization of multimedia databases has raised a number of complex issues for database designers. Some of these issues are [2]:

Modeling: Multimedia information includes media objects, associated metadata, and the objects' temporal and spatial characteristics. This information is continuously manipulated and modified by applications. Some of the different techniques used for modeling multimedia data are [6]:
1. object-oriented modeling: inspired by the object-oriented paradigm, it organizes multimedia information into hierarchical structures in which each multimedia object (e.g., text, audio, video, image) has its set of variables, methods, and messages to which it responds. Such a model can enforce concepts such as encapsulation, data hiding, and multiple inheritance, and can handle the metadata as well. Some of its drawbacks include the difficulty of accessing objects in a collective - rather than individual - manner, the need to handle the database schema independently from the class hierarchy, and the impossibility of creating new objects that are based on portions of existing objects and need only to inherit part of their attributes.
2. temporal models: multimedia objects have associated temporal characteristics, which are particularly important in presentation-type applications. These characteristics specify parameters such as: time instant of an object presentation, duration of presentation, and synchronization among objects in the presentation. Temporal models are called hard when the temporal relationships are specified in a precise manner with exact values for time instants and duration of presentations, or flexible if they allow a range of values to be specified for each time-related parameter.
3. spatial models: multimedia applications are constrained by the size of each window and the window layout. These constraints must be taken into account either in a hard way (assigning specific values for the x and y coordinates of each window corner), or in a flexible way, using difference constraints and specifying relative positions among the various windows in a presentation.
Design: The conceptual, logical, and physical design of multimedia databases remains an area of active research. The general design methodology summarized in Section 3 can still be used as a starting point, but performance and fine-tuning issues are more complex than in conventional databases.
Storage: Storage of multimedia information in conventional magnetic media brings new problems, such as representation, compression/ decompression, mapping to device hierarchies, archiving, and buffering during the I/O operations.
Queries and retrieval: Efficient query formulation, query execution, and optimization for multimedia data is still an open problem and neither query languages nor keyword-based queries have proven to be completely satisfactory.
Performance: While some multimedia applications can tolerate less strict performance constraints (e.g., the maximum time to perform a content-based query on a remote image repository), others are inherently more critical, such as the minimum acceptable frame rate for video playback.

Recent developments in Multimedia Database Systems are expected to bring together two disciplines that have historically been separate: database management and information retrieval. The former assumes a rigid structure for data and derives the meaning of a data instance from the database schema, while the latter is more concerned with modeling the data content, without paying much attention to its structure [2].

There are very few commercial multimedia database management solutions currently available, e.g., MediaWay's Chuckwalla Broadband Media Management System. Nonetheless, many well-known DBMSs support multimedia data types; examples include Oracle 8.0, Sybase, Informix, ODB II, and CA-JASMINE. The way multimedia extensions are handled by each of these systems is ad hoc, and does not take into account interoperability with other products and solutions. It is expected that the MPEG-7 standard will promote the necessary standardization.