A Presentation on

AI and Robotics

Nov 30, 2013 (4 years and 7 months ago)

84 views

A Presentation on

MINING SPATIAL ASSOCIATION RULES IN IMAGE
DATABASES USING 9
-
DLT AND ITS COMPARISON WITH
MINING VIWEPOINT PATTERNS IN IMAGE DATABASES

Submitted By:

KUSUM LATA

College Roll No: 24/CTA/07

University Roll No. 8833

under the Guidance of:

Dr. RAJNI JINDAL

Dept. of Computer Engineering

Delhi College of Engineering, Delhi

Introduction

Many data mining methods have been proposed such as
association rule mining, Sequence pattern mining, Text mining
,Temporal data mining etc.

Because of advances in Information technology vast number
of images have accumulated on the internet, in entertainment,
education and in other multimedia applications

So how to mine spatial patterns from image database has
attracted attention these days.

What is spatial data mining?

I

Spatial

data

mining

refers

to

the

extraction

of

implicit

knowledge

relationships,

or

other

interesting

patterns

stored

in

spatial

databases
.

Applications
:

Geographic

information

system

Architectural

images

Medical

images

Multimedia

information

system

etc
.

Image

representation

Many

image

representation

methods

has

been

proposed

such

as

attribute

relationship

graph

in

which

object

are

represented

by

a

node

and

relation

ship

b/w

nodes

is

represented

by

a

arc
.

Another

representation

is

9
DLT

representation
.

Directional

information

is

one

of

the

most

important

types

of

information

in

an

image

database,

and

the

9
DLT

representation

is

fundamental

in

this

method
.

Here

I

will

discuss

a

DLT

mining

algorithm

to

mine

spatial

association

rule

where

every

image

is

represented

by

a

9
DLT

representation
.

The 9DLT mining Algorithm

The algorithm consists of two phases
.

1. In the first phase, we find all frequent patterns of length one.

2. In the second phase, we use the frequent k
-
patterns,whose lengths are equal to k,
to generate all candidate (k + 1)
-
patterns. For each candidate (k + 1)
-
pattern
generated, we scan the database to count the pattern’s support and check
whether it is frequent. The steps in the second phase are repeated until no
more frequent patterns can be found.

Let I=(i
1
,i
2
……i
w
) be a set of items each of which is an object in an image.
An item set is a subset of I.

A spatial relation between two objects is one of the 9 directional codes
defined in 9DLT representation. A spatial relation is defined on the basis of
this representation.

Example demonstrating how a
symbolic image is transformed
to a 9DLT string:

First the symbolic image is
converted to 9DLT matrix, which
a lower triangular matrix,

Then the 9DLT matrix is
represented in the 9DLT string as
shown in figure.

So (A,B,C,D,5,6,6,7,6,6) is the
9DLT string representation of the
given symbolic image which
consists of two parts: item
(objects) and spatial relations.

Given two frequent 2
-
patterns (A,B,3) AND (A,C,4) as shown in above figure. We
can generate a candidate 3

pattern by joining them. That is we have (A,B,C,3,4,?),
where relation b/w (A,B) and (A,C) are known, but relation b/w (B,C) is unknown .

We need to find possible relations between B and C from (A, B, 3) and (A, C, 4).
The possible relations between (B
,

C), as shown in above figure are 4, 5, and 6. So,
we can obtain three candidate 3
-
patterns:
(A
,
B
,
C
,
3
,
4
,
4
)
,
(
A
,B,C,3,4,
5
), and
(A
,
B
,
C
,
3
,
4
,
6
), where the
underlined relations are the possible relations between B
and C

In General the possible relation
derived by joining two joinable
2 patterns (A,B,X) and (A,B,Y)
to generate a candidate 3
-

pattern (A,B,C,X,Y,Z). The
matrix hence derived is the
complementary matrix.

In this matrix the Z values of an
entry are complementary to
those of its symmetric entry.

For example the Z values of
entry[1,2] are equal to (6,7,8)
and the Z value of its symmetric
entry are [2,1] is equal to (2,3,4)

The Mining Algorithm

The 9DLT
-
Miner Algorithm consists of two phases.

In the First phase ,we find all frequent patterns of length one.

In the second phase, we find use the frequent k
-
patterns (k>=2) to generate
all candidate (k+1)
-
patterns. For each (k+1) pattern generated ,we scan the
database to count its support and check if it frequent. the steps in phase 2
are repeated until no more frequent patterns can be found.

The 9DLT

Miner algorithm consists of one procedure GenCandidates
which is used to generates possible

candidate patterns.

Mining Viewpoint Patterns in Image Databases

Viewpoint patterns refer to patterns that capture the invariant relationships of
one object from the point of view of another object.

These patterns are unique and significant in

images because the absolute
positional information of objects for most images is not important, but rather, it
is the relative distance and orientation of the objects from each other that is
meaningful. Analyzing large set of images, we find that the absolute positional
information of objects do not convey critical perceptual information, but rather,
it is the invariant relationships , in the form of the relative spatial relationships
among the objects in images, that are important.

Above

relationship a
viewpoint pattern
.

Such patterns are insensitive to
translational operations, and in some applications, rotational operations too.

The figure 2 shows the possible 2
-

objects pairs in the plan_1.

Setting the minimum support to 3 , the
frequent 2
-

object patterns generated from
the images in figure 1 are given be

The 2
-
objects table is scanned to determine the frequent 2
-

objects patterns.

Based on 2
-
object table we build the 3
-
object table. For each pair of the object in
the 2
-
object table we adda new object that can be found in the same image to form
a 3
-

object group
.

When scanning a 3
-
object table candidates whose count is greater than or equal
to three are output as 3
-
object viewpoint patterns. This process is repeated until
no more frequent patterns can be found.

A Comparison

As directional information is one of the most important types of information in an
image database, and the 9
-
DLT representation is fundamental in this method. The
9DLT
-
Miner is a spatial mining algorithm, called 9
-
DLT
-
Miner, to mine spatial
association rules in an image database, where every image is represented by the 9
-
DLT representation.

While on the other hand the Viewpoint Miner Algorithm uses another image
representation approach. Viewpoint patterns refer to patterns that capture the
invariant relationships of one object from the point of view of another object.

The viewpoint patterns are unique and significant in images because the absolute
positional information of objects for most images is not important, but rather, it is
the relative distance and orientation of the objects from each other that is
meaningful.. Analyzing large set of images, the absolute positional information of
objects do not convey critical perceptual information, but rather, it is the invariant
relationships , in the form of the relative spatial relationships among the objects in
images is important. Again the Viewpoint Miner is able to mine rotational
-
invariant
patterns while 9
-
DLT not.

17
-
DLT Method

9
-
DLT algorithm is designed to mine for frequent patterns of images represented by
the 9
-
DLT representation, it is extended to mining frequent patterns for images
represented by the other directional representation i.e. 17
-
DLT. The 17
-
DLT is
more direction specific than 9
-
DLT. The complementary matrix for this
representation is derived. For the 17
-
DLT representation, each directional code (not
including 0) in the 9
-
DLT representations divided into two sub
-
directional codes.

B

A

C

F

D

E

(A,B,C,D,E,F,9,3,12,11,13,5,6,5,4,5,3,1
)

CONCLUDING REMARKS AND FUTURE WORK

In this project 9
-
DLT
-
Miner for mining spatial association rules in image databases
discussed in which every image is represented by a
-
9DLT representation. The
algorithm consists of two phases. In the first phase, we find all frequent patterns of
length one. In the second phase, we use the frequent k
-
patterns (k >= 1), whose
lengths are equal to k, to generate all candidate (k + 1)
-
patterns and then scan the
database to count the support and check if a pattern is frequent for each candidate (k
+ 1)
-
pattern generated. The steps in phase 2 are repeated until no more frequent
patterns can be found. Another

algorithm

discussed and

compared with 9
-
DLT is Viewpoint miner.

Viewpoint patterns refer to patterns that capture the invariant relationships of one
object from the point of view of another object. These patterns are unique and
significant in images because the absolute positional information of objects for
most images is not important, but rather, it is the relative distance and orientation of
the objects from each other that is meaningful.

Further 9
-
DLT is extended to 17
-
DLT to mine frequent patterns of images
represented by the 17
-
DLT representation. The complementary matrix for 17
-
DLT
is derived.

There are a number of possible future research directions .We can work for mining
association rules for an image database in which images are represented by other
representations Attribute Relationship Graph.

REFERENCES

Agrawal, R. Srikant, Fast algorithms for mining association rules, in: Proc. of International Conference on
Very Large Data Bases, Santiago
,
Chile, 1994, pp. 487

499

W. Hsu, J. Dai, M. Lee, Mining viewpoint patterns in image databases,

ACM International Conference on
Knowledge Discovery and Data Mining, Washington

DC, 2003, pp. 553

558.

Mining association rules in image databases Anthony J.T. Lee , Ruey
-
Wen Hong, Wei
-
Min Ko,Wen
-
Kwang Tsao,Hsiu
-
Hui Lin, Information Sciences 177,2007, ,pp. 1593

1608.

New algorithms for efficient mining of association rules, Li Shen a, Hong Shen Ling heng a Information
Sciences 118 ,1999, pp. 251
±
268.

Mining frequent patterns in image databases with 9D
-
SPA representation ,Anthony J.T. Lee,Ying
-
Ho Liu,
Hsin
-
Mu Tsai, Hsiu
-
Hui Lin, Huei
-
Wen Wu ,The Journal of Systems and Software 82 ,2009, pp. 603

618.

Michael Hahsler, Wien Bettina, Computational Environment for Mining

Association Rules and Frequent
item Sets, Journal of Statistical software, October 2005.

Y. Rui, T. S. Huang, and S. Chang. Image retrieval: Current techniques, promising

directions and open issues. Journal of Visual Communication and Image

Representation

10
,
1999
.

Ling
-
Yin Wei and Man
-
Kwan, Efficient Mining of Spatial Co
-
orientation Patterns

in

Image database,

2006 IEEE International

Conference on

Systems,

Man, and

Cybernetics
,
Taiwan, October 8
-

11, 2006.

Michael Hahsler, Wien Bettina, Computational Environment for Mining

Association Rules and Frequent
item Sets, Journal of Statistical software, October 2005.

P. Punitha, D.S. Guru, An invariant scheme for exact match retrieval of Symbolic

Images: Information Sciences 26, 2005, pp. 893

907.

Y. Rui, T. S. Huang, and S. Chang. Image retrieval: Current techniques, promising

directions and open issues. Journal of Visual Communication and Image

Representation

10
,
1999
.

Ling
-
Yin Wei and Man
-
Kwan, Efficient Mining of Spatial Co
-
orientation Patterns

in

Image database,

2006 IEEE International

Conference on

Systems,

Man, and

Cybernetics
,
Taiwan, October 8
-

11,
2006.