Database Reference
In-Depth Information
Fig. 4.6
Quadkeys quantization and hashing from GPS, and images ground distance estimation
using Microsoft Bing Map service
4.2.5
GPS Context-Based Filtering
Context information collected by mobile sensors plays an important role to help to
identify users' visual intents. As Fig.
4.5
illustrates, similar with the inverted file
index method, each piece of image context information is indexed with the image
itself during the off-line database construction.
In our system, GPS information from sensors is utilized and associated with
each image taken by the phone camera. A filter-based process is used to remove
the non-correlated images after the initial visual search. This is because GPS as
an important context filter can be used to efficiently explore users' true intents by
precisely knowing their locations. This process is formulated as:
S
L
(
q
,
d
)=
s
(
q
,
d
)
·
ˆ
(
q, d
)
1
,
if
dist
quadkey
(
q
,
d
)
∈
Q
where
ˆ
(
q, d
)=
(4.7)
0
,
if
dist
quadkey
(
q
,
d
)
∈
Q
.
This filter is based on the GPS effective region Q, which describes the geographical
distance between the query and the database images. We defined
dist
quadkey
(
The visual similarity term
s
(
q
,
d
)
is modulated by a location-based filter
ˆ
(
q, d
)
q
,
d
)
as
the quadkey distance between the query
q
and the database image
d
.
Search WWH ::
Custom Search