25 Multimedia Database
Dr R. Baskaran
OVERVIEW
Recent developments in information systems technologies have resulted in computerizing many applications in various business areas. Data has become a critical resource in many organizations, therefore efficient access to data, sharing or extracting information from the data, and making use of this information have become urgent needs. As a result, many efforts to integrate the various data sources scattered across several sites as well as extract information from these databases in the form of patterns and trends. These data sources may be databases managed by database management systems, or they could be data warehoused in a repository from multiple data sources. The advent of the World Wide Web (WWW) in the mid 1990s has resulted in even greater demand for managing data, information, and knowledge effectively. There is now so much data on the Web that managing it with conventional tools is becoming almost impossible. New tools and techniques are needed to effectively manage these data. Therefore, various tools are being developed to provide interoperability and warehousing between multiple data sources and systems, as well as to extract information from the databases and warehouses on the Web. Data in Web databases are both structured and unstructured. Structured databases include those that have some structure such as relational and object databases. Unstructured databases include those that have very little structure such as text, image, audio, and video databases. In general, multimedia databases are unstructured. Some text databases are semi-structured databases, meaning that they have partial structure. The developments in multimedia database management systems have exploded during the past decade. While numerous papers and some texts have appeared in multimedia databases, more recently these databases are being mined to extract useful information. Furthermore, multimedia databases are being accessed on the Web. That is, there is currently little information about providing a complete set of services for multimedia databases. These services include managing, mining, and integrating multimedia databases on the Web for an electronic enterprise. The focus of this book is on managing and mining multimedia databases for the electronic enterprise. We focus on database management system techniques for text, image, audio, and video databases. We then address issues and challenges regarding mining the multimedia databases to extract information that was previously unknown. Finally, we discuss the directions and challenges of integrating multimedia databases for the Web. In particular, e-business and its relationship to managing and mining multimedia databases will be discussed. As mentioned earlier, there are hardly any texts on providing a comprehensive set of services for multimedia data management, although numerous research papers have been published on this topic. The purpose of this book is to discuss complex ideas in multimedia data management and mining in a way that can be understood by someone who wants background information in this area. Technical managers as well as those interested in technology will benefit from this book. We employ a data-centric approach to describe multimedia technologies. The concepts are explained using e-commerce and the Web as an application area.
MULTIMEDIA DATABASE MANAGEMENT
A multimedia database system is comprised of a multimedia database management system (MM-DBMS) that manages a multimedia database, which is a database containing multimedia data. Multimedia data may include structured data as well as semi structured and unstructured data such as voice, video, text, and images. That is, an MM-DBMS provides support for storing, manipulating, and retrieving multimedia data from a multimedia database. In a certain sense, a multimedia database system is a type of heterogeneous database system because it manages heterogeneous data types.
An MM-DBMS must provide support for typical database management system functions. These include query processing, update processing, transaction management, storage management, metadata management, security, and integrity. In addition, in many cases, the various types of data such as voice and video have to be synchronized for display, and, therefore, real-time processing is also a major issue in an MM-DBMS.
MM-DBMSs are becoming popular for various applications including C4I, CAD/CAM, air traffic control, and, particularly, entertainment. While the terms multimedia and hypermedia are often used interchangeably, we differentiate between the two. While an MM-DBMS manages a multimedia database, a hypermedia DBMS not only manages a multimedia database, but also provides support for browsing the database by following links. That is, a hypermedia DBMS contains an MM-DBMS. Recently, there has been much research on designing and developing MMDBMSs, and, as a result, prototypes and some commercial products are now available.3,4,21,61-63,92,102,129 However, as stated by Dao and Thuraisingham,124 there are several areas that need further work. Research on developing an appropriate data model to support data types such as video is needed. Some experts have proposed object-oriented database management systems (OO-DBMS) for storing and managing multimedia data because they have been found to be more suitable for handling large objects and multimedia data such as sound and video which consume considerable storage space.139 Although such systems show some promise, they are not sufficient to capture all of the requirements of multimedia applications. For example, in many cases, voice and video data which may be stored in objects have to be synchronized when displayed. The constraints for synchronization are not specified in the object models. Another area that needs research is the development of efficient techniques for indexing. Data manipulation operations such as video editing are still in the early stages. Furthermore, the multimedia databases need to be integrated for many applications as they are distributed.
Some Important Definitions
Media Types and Multimedia
Media refer to the types of information or types of information representation, such as alphanumeric data, images, audio, and video. There are many ways to classify media. Common classifications are based on physical formats and media relationships with time.
Static media do not have a time dimension, and their contents and meanings do not depend on the presentation time. Static media include alphanumeric data, graphics, and still images. Dynamic media have time dimensions, and their meanings and correctness depend on the rate at which they are presented.
Dynamic media include animation, audio, and video. These media have their intrinsic unit intervals or rates. For example, to convey a perceptually smooth movement, video must be played back at 25 frames per second (or 30 frames, depending on the video system used).
Multimedia refers to a collection of media types used together. It is implied that at least one media type is not alphanumeric data (i.e., at least one media type is image, audio, or video). In this book, ”multimedia” is used as an adjective—so we will specifically say multimedia information, multimedia data, multimedia system, multimedia communications, multimedia applications, and so forth. Multimedia data refers to the computer-readable representation of multiple media types. Multimedia information refers to the information conveyed by multiple media types. Sometimes, multimedia information and multimedia data are used interchangeably.
We sometimes use multimedia or media item and object to refer to any autonomous entity in an MIRS that can be queried, retrieved, and presented. The term “object” may not be properly defined in the technical object-oriented (OO) sense. The context should make it clear whether it is used in a general sense or refers to a properly defined object in an OO approach.
Multimedia Indexing and Retrieval
DBMSs retrieve items based on structured data using exact matching. IR is also called text-based retrieval. Content-based retrieval refers to retrieval based on actual media features such as color and shape, instead of text annotation of the media item. Content-based retrieval is normally based on similarity instead of an exact match between a query and a set of database items. MIRS refers to a basic system providing multimedia information retrieval using a combination of DBMS, IR, and content-based retrieval techniques. In an MIRS, some issues such as versioning and security control may not be fully implemented. A fully fledged MIRS is called a multimedia DBMS (MMDBMS).
Feature Extraction, Content Representation, and Indexing
In MIRSs, one of the most important issues is feature extraction or content representation (what are the main features or contents in a multimedia item). Feature extraction may be an automatic or semiautomatic process. In some of the content based retrieval literature, feature extraction is also called indexing. When the term “index” is used as a noun, it refers to a data structure or to the organization of extracted features for efficient search and retrieval.
DBMSs and Their Role in Handling Multimedia Data DBMSs are now well developed and used widely for structured data. The dominant DBMSs are relational database management systems (RDBMSs). In RDBMSs, information is organized in tables or relations. The rows of the table correspond to information item or records, while the columns correspond to attributes. The structured query language (SQL) is used to create such tables and to insert and retrieve information from them. We use a simple example to show how to use SQL to create a table and insert and retrieve information from it. Suppose we want to create a table containing student records consisting of the student number, name, and address. The following statement is used:
- create table STUDENT ( stu# integer, name char(20), address char(100) );
- When we want to insert student records into the table, we use the SQL insert command as follows:
- insert into STUDENT values (10, “Kannan, Arputharaj”, “2 Main St., Quarters”, “Chennai”);
- Information in the table is retrieved using the SQL select command. For example, if we want to retrieve the name of a student with student number 32, we use the following query statement:
- select name from STUDENT where stu#=32
- Attributes in a RDBMS have fixed types with fixed widths. In the above example, the attribute stu# is an integer type of fixed length of 32 bits. Thus RDBMSs are well suited for handling numeric data and short alphanumeric strings.
To support large variable fields in a RDBMS, a concept called binary large objects (BLOBs) was introduced. A BLOB is a large bit string of variable length. For example, if we want to store students’ pictures in the above student record example, we can create a table using the following statement:create table STUDENT ( stu# integer, name char(20), address char(100) picture BLOB);BLOBs are normally just bit strings and operations such as comparison can not be carried out on them. That is, a RDBMS does not know the contents or semantics of a BLOB. All it knows is a block of data.
The main difference between the BLOB and the object is that the object is properly defined, including its properties and allowed operations on the properties, while the BLOB is not.
The concepts of BLOBs and objects are a step toward handling multimedia data. But BLOBs are used just to store large data. While objects contain some simple attributes, mANY more capabilities should be developed to handle contentbased multimedia retrieval. Some of the required capabilities are as follows:
• Tools, to automatically, or semiautomatically extract contents and features contained in multimedia data;
• Multidimensional indexing structures, to handle multimedia feature vectors;
• Similarity metrics, for multimedia retrieval instead of exact match;
• Storage subsystems, redesigned to cope with the requirements of large size and high bandwidth and meet realtime requirements;
• The user interface, designed to allow flexible queries in different media types and provide multimedia presentations.
Integrated Approach to Multimedia Information Indexing and Retrieval
From the above discussion we see that DBMSs and IR cannot fully meet the requirements of multimedia indexing and retrieval, so new techniques to handle special characteristics of multimedia data are required. Nevertheless, we recognize that DBMSs and IR can play important roles in MMDBMSs. Parts of multimedia data, such as the creation date and author of a multimedia document, are structured. This structured data can be handled with DBMS techniques. Text annotation is still a powerful method for capturing the contents of multimedia data, so IR techniques have an important role to play. To summarize, an integrated approach combining DBMSs, IR, and specific techniques for handling multimedia data is required to develop an efficient and effective MIRS.