Database Reference
In-Depth Information
Algorithm 23.
SELECT * FROM Landscapes L1, Landscapes L2
WHERE Manhattan_dist(L1.Picture_Texture, L2.Picture_Texture) <= 0.03
Algorithm 24.
SELECT * FROM Landscapes
WHERE Place = 'Paris' AND
Manhattan_dist(Figure_Texture, Example) <= 0.03 AND
Manhattan_knn(Figure_Texture, Example) <= 5;
FMI-SiR
O
allows the query processor to choose the best execution strategy regarding all query con-
ditions. Therefore, the most selective condition can be used to filter the results or intersections can be
employed between the individual index results before accessing the data blocks. However, optimizer
cannot detect special similarity constructions, as a Range and a k-NN query with the same query center.
This happens because only the built-in rewriting rules of the query processor are evaluated. To overcome
this issue, it would be necessary to modify the DBMS query processor. Another alternative would be to
combine the two approaches described in this section: a blade that intercepts SQL instructions and rewrite
them in the best query plan using the standard DBMS resources as well as those provided by a module
for similarity searching. However, this is a solution that only can be implemented changing the code of
the underlying DBMS, and thus only can be implemented in a way specific for a given product.
CONCLUSION
Most of the systems available to query multimedia data by similarity were developed considering a
specific data domain and presenting a closed architecture. This approach does not allow applications
to extend the access over traditional data in order to also deal with other data types, such as images
and audio. This can be a big issue, for example, for the extension of typical applications of a medical
information system, such as the electronic patient record system, in a way that they could also support
image retrieval by similarity to support decision making.
Commercial products, such as Oracle InterMedia and IBM DB2 AIV Extenders, support the man-
agement of several types of multimedia data through user-defined functions and types. Although this
approach can use the existing highly optimized algorithms for each specific similarity operation, it does
not allow optimizations among the operators nor their integration with the other operators used in a query.
Therefore, it is fundamental that DBMS support the management of multimedia data by similarity
through native predicates in SQL, built into an architecture that is capable to be easily adjusted to the
particular needs of each application domain. This chapter contributes to support similarity queries as a
built-in resource in relational DBMS, addressing the fundamental aspects related to the representation
