Databases Reference
In-Depth Information
2.4.2
Dynamic metric
We introduce a dynamic metric between the image data items, according to a
context. Since each image data item can be represented as a vector via the union
operator
defined in Section 4.2, we can utilize the metric, which we defined for
two distinct words in
4,
5,
7)
, to compute the similarity between metadata items of
images. The dynamic metric
⊕
ρ
(
x, y;
s
l
) for
x, y
ε
MDS
is introduced to compute the
similarity between metadata items for images.
The metric
(
x, y;
s
l
) to compute the similarity between metadata items
x, y
of
two images in the given context
s
l
is defined as follows:
ρ
This metric, because of the presence of dynamic weights
c
j
's, can faithfully reflect
the change of the context.
3
Semantic Associative Search for Metadata for Images
The proposed system realizes the semantic associative search for metadata items
for images.
The basic function of the semantic associative search is provided for context-
dependent interpretation. This function performs the selection of the semantic
subspace from the metadata space. When a sequence
s
l
of context words for
determining a context is given to the system, the selection of the semantic subspace
is performed. This selection corresponds to the recognition of the context, which
is defined by the given context words. The selected semantic subspace corresponds
to a given context. The metadata item for the most correlated image to the context
in the selected semantic subspace is extracted from the specified image data item
set
W
. By using the function defined in Section 2.4.2, the semantic associative
search is performed by the following procedure:
1.
When a sequence
s
l
of the context words for determining a context (the
user's impression and the image's contents) are given, the Fourier expansion
is computed for each context word, and the Fourier coefficients of these
words with respect to each semantic element are obtained. This corresponds
to seeking the correlation between each context word and each semantic
element.
2.
The values of the Fourier coefficients for each semantic element are summed
up to find the correlation between the given context words and each semantic
element.
3.
If the sum obtained in the step 2 in terms of each semantic element is greater
than a given threshold
ε
s
,
the semantic element is employed to form the seman
