The document discusses using spatial semantics and association rules to retrieve ocean images based on salinity and temperature variations over time. It proposes using concentric circles to annotate location information for objects in images and mining inter-transaction association rules to predict future trends. Experimental results demonstrate retrieving additional images based on salinity and temperature patterns to help users forecast variations.
Social media recommendation based on people and tags (final)
Extracting ocean
1. Extracting Spatial Semantics in
Association Rules for
Ocean Image Retrieval
Author:Li-Jen Kao,Yo-Ping Huang, Frode Eika Sandnes
Source:Proceedings of the 2009 IEEE International Conference on Systems,
Man, and Cybernetics
San Antonio, TX, USA - October 2009
Advisor:Yin-Fu Huang
Student:YU-HSIEN CHO
2. OUTLINE
I. Introduction
II. Related Works
1. CBIR
2. Inter-transaction Association Rules
Algorithm
III. Ocean Image Retrieval
IV. Experimental Results And
Discussion
V. Conclusion
3. I. Introduction
Database:Argo is a global research organization that
deploys over 3,000 freedrifting profiling floats that measure
the temperature and salinity of the upper 2000 meters of the
ocean. This facilitates continuous monitoring of the
temperature and salinity of the upper ocean.
In order to get effective retrieval results, the image retrieval
technology is critical to image database systems.
The inter-transaction association rules algorithm is applied to
get salinity or temperature variation patterns.
And use Euclidean distance to measure the distance between
these two objects
4. II. Related Works
1. CBIR-Content-based Image Retrieval
Modern multimedia technologies have led to huge and
growing archives of images in diverse applications such as
medicine, remote-sensing, entertainment, education and
online information services.
In recent years, automatic indexing and retrieval based on
image content has become more desirable for developing
large volume image retrieval applications.
(Fig. 1)
*Color, shape, texture and Spatial Semantic are the main
features both humans and computers used to recognize
images.
6. QBIC:Query By Image Content
IBM QBIC (Query by Image Content) system is the first capture
system, can be retrieved, the conditions include keywords and
images in color, texture, shape and other characteristics.
Provides three ways to search:
First, the user can specify the main color of the query image.
Second, query the probability of relative pixel from the
image.
Third, drawn by the users’ expectations of image which needed
the color of object box, and set the shape of their
representatives.
7. VisualSEEK:
VisualSEEKis a allows users to search image and video on
the World Wide Web, developed by Columbia University's
New Media Technology Center.
The retrieval feature contains color, shape, texture and
spatial information, spatial relationships (Fig. 2)between
objects is also taken into consideration in order to improve
the ability of search.
8. Fig.2 The reference model with
concentric model Fig. 3 Salinity variations occurred in
sub-zone K1, K2, L1 and L2.
Table. 1The radius and angle to x-axis for
each sub-zone in
the concentric circles.
9. 2. Inter-transaction Association Rules
Algorithm
In order to minimize the effort involved in mining uninteresting
rules, a sliding window denoted by w is introduced.
10.
11. Example:
Fig. 4 shows a transaction database T
with five transactions located at intervals
1, 4, 6, 9, and 10. And w=4 .
Assuming that w has a value of 4, we
will have five sliding windows W1, W2,
W3, W4, and W5, with addresses of 1, 4,
6, 9, and 10, respectively.
From the figure, it can be seen that
subwindow W1[0] contains items a, b, e,
g, and j while subwindow W1[3] contains
c, f, and j.
W1 will be
{a(0),b(0),e(0),g(0),j(0),c(3),f(3),j(3)}
Fig.4 Transaction database
and sliding widows.
12. Example:
Let minsup and minconf be 40 percent and 60 percent
respectively. Then, an example of an intertransaction
association rule which will be mined from the database in
Fig. 4 will be:
This rule has a support of 40 percent and a confidence
of 66 percent.
13. III. Ocean Image Retrieval
In this section, the process of how to retrieve the ocean image
is introduced.
Before retrieving images in the database, the objects are cut
from query image according to their color in ocean image and
a set of concentric circles is also applied to the query image to
help annotate location information to objects cut from query
image.
The following method is used to identify the location of an
object.
The object location is denoted by its coordinate
15. The following example explains how to get additional images to
predict future variation trends.
Example: Assume there are total n objects {O1,O2,…….,On} cut fro
the query image.
16.
17. IV. Experimental Results And
Discussion
This study utilized ocean salinity and temperature image at
depth of 100 meters from January 2001 to December 2006.
A database was built comprising these images.
Table 2 shows the market-basket type transaction set of
salinity and temperature variations. Each transaction is
recorded with spatial and temporal information.
Table 3 shows the intervals that the salinity and
temperature variations are mapped to. NOR events are
used to classify situations where no variations occur in a
specific sub-area.
Table 4 shows the results of the quantitative attributes in
Table 2 mapped to several intervals.
18. Table. 2 A database with quantitative attributes
“Salinity variation” and “Temperature variation”
Table. 3 The mapping intervals for
salinity and temperature quantitative values.
19. Table. 4 The “Salinity variation” and “Temperature variation”
Mapping results from Table. 2.
Table 5 shows part of the discovered inter-transaction association
rules. For example, the rule, H1SRL(0) -> A2SRL(3), means that if
salinity rose from 0.05 psu to 0.15 psu in the area that is in the west-
south sector near Taiwan, then salinity will rise from 0.05 psu to 0.15
psu in the east sector far away from Taiwan the next 3 months.
Table. 5 Partial discovered
inter-transaction association rules.
21. V. Conclusion
This study proposes a strategy where concentric circles are
used as a reference model to annotate location information
to the objects in image.
This allows the spatial semantics to be maintained for either
the images in database or the query image.
The proposed intelligent image retrieval system can get more
images to help user predict future salinity or temperature
variation trends.