Database Reference
In-Depth Information
of similarity queries in SQL. Having this goal in mind, the solutions for similarity query representation
and execution presented herein have several interesting characteristics.
First, the SQL extension presented enables representing similarity queries as just one more type of
predicate, leading to the integration of similarity as operations in relational algebra. This characteristic
enables extending the optimizers of the relational DBMS to treat and optimize similarity queries as well.
Second, the presented retrieval engines shows how to benefit from improvements on data retrieval
techniques aimed at similarity, such as techniques involved in the similarity evaluation and index struc-
tures that support similarity operators.
Third, the presented solutions can act as a hub for the development of algorithms to perform broadly
employed similarity operations regarding data analysis. For example, data mining processes often require
performing similarity operations, and having them integrated in the database server, possibly optimized
by a MAM, can be feasible in the future.
FUTURE RESEARCH DIRECTIONS
There is a diversity of application areas for similarity retrieval systems, including: medicine, education,
entertainment and others, each of which usually presenting different requirements. We observed that
important requirements for several applications, such as getting a timely answer for a similarity query,
have been widely explored in many works. However, many other topics regarding representation, index-
ing and searching of multimedia data are still open issues to be explored.
With concern to the retrieval quality, approaches that aims at improving the semantics of multimedia
queries are highly desired. Current techniques yet require much improvement to satisfy the requirements
of specialized applications. The chapter presented several methods that represent a starting point to do
this, such as the multiple center queries, which are a straightforward way to support relevance feedback
requirements.
Regarding the retrieval efficiency, existing structures and algorithms suffer from the
dimensionality
curse
and the search space deserves to be mapped into simpler and more relevant subspaces. Moreover,
algorithms for executing similarity operations are worth improving, such as those for aggregate similar-
ity and similarity joins.
Currently, another issue to be tackled is the lack of flexibility of the available system architectures
to query multimedia data by similarity. As the techniques for inferring the content of complex data and
the process of similarity evaluation evolve, each system for multimedia retrieval tends more and more
to rely on particularized solutions. This panorama can lead to interoperability problems as well as to
redundant development efforts. Therefore, it is necessary to conceive software architectures to manage
multimedia data in such a way that base functionalities and structures can be encapsulated in a manner
equivalent to those that have made DBMS so successful in the past 40 years.
REFERENCES
Baluja, S., & Covell, M. (2008). Waveprint: Efficient wavelet-based audio fingerprinting.
Pattern Rec-
ognition
,
41
(11), 3467-3480. doi:10.1016/j.patcog.2008.05.006
