Databases Reference
In-Depth Information
metadata items are used as the basic words in the English dictionary
“Longman Dictionary of Contemporary English
12)
.” Each metadata item
for space creation corresponds to a row of the matrix
M
. In the setting of a
row of the matrix
M,
each column corresponding to the features which
appear in the explanation of the data item is set to the value “1”. If the
feature is used as a negative meaning, the column corresponding to it is set
to the value “-1”. And, the other columns are set to the value “0”. This
process is performed for each of 2000 metadata items. And then, each
column of the matrix is normalized by the 2-norm to create the matrix
M
.
2.
By using this matrix
M,
the
MDS
is created as the orthogonal space. The
creation method of the orthogonal space is described in Section 3.1. This
space represents the semantic space for computing contexts and meanings
of the metadata items.
To automatically create the data matrix
M
from the dictionary, several filters are
used, which remove unnecessary elements (words), such as articles and pronouns,
and transform conjugations and inflections of words to the infinitives. Those
elements are removed from the features characterizing each data item. The
unnecessary words are not used as features in the data matrix
M
.
(filter-1) This filter eliminates the unnecessary elements, such as articles and
pronouns.
(filter-2) This filter transforms conjugations and inflections to the infinitives.
(filter-3) This filter transforms the large characters to the small ones.
(filter-4) This filter transforms clipped form words to the corresponding original
words.
(filter-5) The rows of the matrix
M
are created for each data item by using the
filtered features which characterize the data item.
Each metadata item (metadata item for images, metadata item for context
representation) is mapped into the metadata space, by computing the Fourier
expansion for the
n
-dimensional vector representing the metadata item itself. These
metadata items are defined as metadata by using the
n
features. These metadata
items are used as context words and metadata items for retrieval candidate images.
2
Creation of a Metadata Space and Basic Functions
In this section, we introduce a creation method of the metadata space
MDS
for
systematically storing metadata and for implementing the semantic associative
search for images
4,
5)
.
