26 Multimedia Databases
Dr R. Baskaran
MULTIMEDIA DATABASE SYSTEMS
INTROUCTION
Various architectures are being examined to design and develop a multimedia database management system (MM-DBMS). These architectures fall under different categories, and this chapter examines the various types of architectures. One architecture type involves integrating multimedia data with the database system. There are two approaches. In the loose coupling approach, the multimedia data is managed by the file system, while the database system manages the metadata. In the tight coupling approach, the multimedia data is managed by the database system. Another type of architecture is schema architecture. For example, does the three-schema architecture apply for a multimedia database system? A third type of architecture is functional architecture, describing the functions of a multimedia database system. A fourth type of architecture is whether a multimedia database system extends a traditional database system. This is what we call a system architecture. A fifth type of architecture is a distributed architecture, where a multimedia database is distributed. Finally, multimedia databases may be heterogeneous in nature and need to be integrated. The architecture for integrating heterogeneous databases is known as interoperable architecture.
LOOSE COUPLING VERSUS TIGHT COUPLING
This section describes the loose coupling versus tight coupling approaches to designing a multimedia database system. In the loose coupling approach, the DBMS is used to manage only the metadata, and a multimedia file manager is used to manage the multimedia data. Then there is a module for integrating the DBMS and the multimedia file manager. Metadata, the multimedia file manager, and the module for integrating the two. The advantage of the loose coupling approach is that one can use various multimedia file systems to manage the multimedia data.
In tight coupling architecture, the DBMS manages both the multimedia database and the metadata. That is, the DBMS is an MM-DBMS. Tight coupling architecture is advantageous because all DBMS functions can be applied on the multimedia database. This includes query management, transaction processing, metadata management, storage management, and security and integrity management.
Note that with the loose coupling approach, unless the file manager performs the DBMS functions, the DBMS only manages the metadata for the multimedia data. Much of the discussion in this book assumes a tight coupling design. That is, the MM-DBMS manages the multimedia database and performs various functions such as query processing and storage management.
SCHEMA ARCHITECTURE
Schema architectures can be described in various ways with respect to different characteristics. Schema is essentially the metadata that describes the multimedia data. Here, the external schema will define the views that users have of the database, such as video or audio views. The logical schema is based on the data model for the multimedia database. One can also look at schema from another point of view. Instead of multimedia data, assume that individual data types are stored in separate databases. For example, video schema will describe the video database and audio schema will describe the audio database.
DATA MODELING
In representing multimedia data, several features have to be supported. First of all, there has to be some way to capture the complex data types and all the relationships between the data. Various temporal constructs such as play-before, play-after, and play-together must be captured. a representation of a multimedia database. In that representation, there are two objects, A and B. Object A consists of 2000 frames, and Object B consists of 3000 frames. Object A consists of a time interval between 4/95 and 8/95, and Object B consists of a time interval between 5/95 and 10/95.
An appropriate data model is critical for representing a multimedia database. Relational, object-oriented, and object- relational data models have been examined to represent multimedia data. Some argue that relational models are better because they can capture relationships, while others argue that object models are better because they represent complex structures. Still others argue that hyper semantic data models are better because they capture both objects, relationships, and rules.
TYPES OF METADATA
Metadata may also include combinations of data types such as audio and video or text and video. Metadata could also be characterized by whether it depends on the context or not. For example, in the case of text metadata, content-dependent metadata may include information that depends on the content, e.g., the keywords in the story are vehicle, employee, and license. Content-independent metadata may include information such as, “a book has chapters and paragraphs.” Metadata may also be extracted from multimedia data and used to understand the multimedia data.
MULTIMEDIA QUERY PROCESSING
All of the database functions for an MM-DBMS are more complex because the data may either be structured or unstructured for multimedia databases. Furthermore, handling various data types such as audio and video is quite complex. In addition to these basic DBMS functions, an MM-DBMS must also support real-time processing to synchronize multimedia data types such as audio and video. Quality of service is an important aspect for an MM-DBMS. For example, in certain cases, high quality resolution for images may not always be necessary. Special user interfaces and multi-modal interfaces are also needed to support different media.
DATA MANIPULATION FOR MULTIMEDIA DATABASES
While query processing is the major focus of this chapter, data manipulation, which is much broader than query processing and includes editing as well as transaction management, is discussed here. Therefore, we first provide an overview of data manipulation and then discuss query processing. As mentioned earlier, data manipulation involves various aspects. Support for querying, browsing, and filtering the data is essential for multimedia databases. In addition to querying the data, one may also want to edit the data. Query processing.
Object Editing
Consider editing multimedia objects. For example, two objects may be merged to form a third object. One can project an object to form a smaller object. As an example, objects may be merged based on time intervals, and an object may be projected based on time intervals. Objects may also be updated in whole or in part.
Browsing
Browsing multimedia data is essentially carried out by a hypermedia database management system. The multimedia data is presented in terms of nodes and links. One traverses the links to reach the nodes and clicks on the links to get the relevant multimedia data.
Filtering
Filtering is the process of removing unnecessary material from data. This occurs quite often in video data where material inappropriate for children may be removed from a video clip. This means that the video clips have to be filtered and the filtered data displayed to the users.
Transaction Management
This function is also an aspect of data manipulation as it involves querying and updating databases. There has been some discussion as to whether transaction management is needed in MM-DBMSs. We feel that this is important because, in many cases, annotations may be associated with multimedia objects. For example, if one updates an image, its annotation must also be updated. Therefore, the two operations have to be carried out as part of a transaction. Unlike query processing, transaction management in an MM-DBMS is still a new area. Associated with transaction management are concurrency control and recovery. The issue is, what are the transaction models? Are there special concurrency control and recovery mechanisms? Much research is needed in this area.
Update Processing
Update processing is usually considered part of query processing or transaction management. Update processing is essentially updating the multimedia data and is often a single user update. An example of a request is, “update the text paragraph C in document A to paragraph B.” There have been many discussions on updating video and audio data. That is, how do you update part of a video or parts of an image? Is it possible to delete part of an image and replace it with some other image? We do not have satisfactory answers to update processing for multimedia databases. This is also related to the difficulties in transactions processing in multimedia databases.
MULTIMEDIA INDEXING
The storage manager is responsible for accessing the database. To improve the efficiency of query and update algorithms, appropriate access methods and index strategies have to be enforced. That is, in generating strategies for executing query and update requests, the access methods and index strategies that are used need to be taken into consideration. The access methods used to access the database depend on the indexing methods. Therefore, creating and maintaining appropriate index files is a major issue in database management systems. By using an appropriate indexing mechanism, the query processing algorithms may not have to search the entire database. Instead, the data to be retrieved could be accessed directly. Consequently, the retrieval algorithms are more efficient.
Extensive research has been performed to develop appropriate access methods and index strategies for relational database systems. Some examples of index strategies are B trees and hashing.The major issues in storage management for multimedia databases include developing special index methods and access strategies for multimedia data types.Content-based data access is important for many multimedia applications.However,efficient techniques for content-based data access are still a challenge.
One could also develop indexes for the annotations. Since the annotations describe information about the multimedia data, if one can access the annotations, one can get the multimedia data. Various indexing techniques for multimedia data have been proposed. Also, extensions to B trees and B + trees have been proposed for multimedia data.
Other storage issues include caching data. How often should data be cached? Are there any special considerations for multimedia data? Are there special algorithms? Also, storage techniques for integrating different data types are needed. For example, a multimedia database may contain video, audio, and text databases instead of just one data type. The displays of these different data types have to be synchronized. Appropriate storage mechanisms are needed so that there is continuous display of the data. It is important that the display of multimedia data is synchronized with the retrieval of the data. This is especially true in the case of video on demand (VOD). Suppose we want to look at a film and we retrieve the video through the Internet or with the VOD boxes from our television sets. If the presentation is much quicker than the retrieval, there will be periods where we will have no display, which may not be acceptable to many viewers. In other words, we need continuous presentation of the film. In some cases, we can cache the film in order to get continuous display. It is impossible to cache all the films, and, therefore, we need efficient synchronization techniques. Typically, a user may specify the quality of service primitives, and the video should be presented according to the specifications. Video streaming has been a topic of much recent research. Furthermore, we now have special devices that consumers can purchase and attach to their television sets so that quality of service video on demand is and possible.