This preview shows page 1. Sign up to view the full content.
Unformatted text preview: n
the database for future reference. It may either replace the old version or exist as
another version of the data. If users constantly create new versions of data, the
database size will grow rapidly. To reduce the need to store new versions of data,
a multimedia database management system stores the operations to generate the
new data (called derived data) rather than storing the new data itself.
Support for Multimedia Data Query Some desired information can be extracted from the data stored in a database by
initiating queries. Queries contain predicates that must be satisfied by any data
retrieved. It is easy to formulate predicates and look for its matching data in
conventional databases. For example, "Find all employees for whom CITY =
"PUNE" and AGE > 40". However, this is not an easy task while dealing with
multimedia data. The three commonly used methods in existing multimedia
database systems for querying multimedia data are:
1. Manual keyword indexing. This method is currently the most popular and
straightforward method to query multimedia databases. In this method, descriptive
keywords are manually associated with every multimedia data stored in the
database. For example, keywords like “long face, sharp pointed nose, broad
moustache, no beard, fair complexion, mole on right side of chin” may be used to
describe an image data of a face. These keywords are metadata, since they are data
about multimedia data. For querying, a user specifies some keywords and the
database management system looks for images with matching keywords (the
images themselves are not queried). All images with matching keywords are
returned as the result of the query.
This method has the following limitations:
- Keyword classification is subjective since it is performed by a human.
- Exceptions will always exist, and some data may be incorrectly classified.
- Keywording is usually limited to a well-defined abstraction of the data (for
example, for every image of a face, a specific set of features is classified). This
means that if the abstraction becomes altered, then all...
View Full Document
- Spring '14