Databases Reference
In-Depth Information
For example, the color names defined by ISCC and NBS can be used as the
features. The creation of the metadata for images can be done automatically, as
the color of the image in the CIE color system can be obtained easily and can be
transformed to the Munsell color system. The creation of the metadata can be
done automatically.
However, the difficulty in automatic extraction is how to correspond the colors
with explanatory keywords, that is, how to create the metadata for keywords. To
solve this difficulty, the results of psychological experiments can be used, because
many word association tests had been done on the relation between colors and
psychological effects, e.g. showing a single color and asking for the reminded
words.
7
Conclusion
In this chapter, we have reviewed several database systems dealing with Kansei
information for extracting media data according to the user's impression and the
image's contents. Those system provide functions for defining and retrieving images
with Kansei information. Those functions are realized by computing correlations
between Kansei information of media data and a user's request represented with
Kansei information.
In this field, the development for learning mechanisms is important for
supporting adaptivity to individual variations in Kansei. The implementation of
automatic extraction mechanisms for Kansei information is also very important
to realize actual Kansei database system environments.
We have introduced new methodology for retrieving image data according to
the user's impression and the image's contents. We have presented functions and
metadata for performing semantic associative search for images. The functions
are realized on the basis of mathematical model of meaning.
For the creation of the metadata for images, we have introduced three methods
(Method-1, Method-2 and Method-3). The metadata created by those methods is
categorized in the type of
content-descriptive metadata
according to the metadata
classification for digital media presented in
2)
. Furthermore, the metadata created
by the first two methods, Method-1 and Method-2, is categorized into the
contentdescriptive domain-dependent metadata, and the metadata by the third
method is classified as the type of content-descriptive domain-independent
metadata.
We have implemented the semantic associative search system to clarify its feasibility
and effectiveness. Currently, we are designing a learning mechanism to adapt metadata
for context representation and images according to the individual variation. The
learning is a significant mechanism for semantic associative search, because the
judgment of accuracy for the retrieval results might be dependent on individuals.
We will use this system for realizing a multimedia metadatabase environment.
As our future work, we will extend this system to support multimedia data retrieval
