Combination best combination terms. So make of

Combination
Terms Using Genetic Algorithm for Image Retrieval

T.Thamizhvendhan

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!


order now

Assistant Professor

Department
of Computer Science, ES Arts and Science college,

Villupuram

 

Abstract— Image
in web documents is worth a thousand words. The meaning of web image is highly
individual and subjective. So it an essential to extract semantic information
from associated with web page images. There are number of techniques to extract
the semantic information are textual keywords about the web page images but
which could not able to associated with web page images. In this research work
a strength matrix is being proposed which combines the evidence extracted from
text and visual content of web page images. The strength matrix is based on
frequency occurrence of keywords and the textual information pertaining to web
page images. The strength matrix is created by document crawler and the genetic
algorithm takes input as keywords from the strength matrix and gives an output
as best combination terms. The best combination terms is given to image
retrieval system for building index for best combination terms. So make of this
index we can improve precision of retrieval.

Keywords— Binary strength
matrix, image retrieval, genetic algorithm, combination terms.

INTRODUCTION

Today massive growth of World Wide
Web, people are gaining access to large amount of information. However locating needed
and relevant information is very difficult. Current search engines have
achieved certain degree of success to retrieve text documents. Retrieving
images in the World Wide Web is challenging issue for Image search engines
example GOOGLE, CORBIS. The approach gave results are low precision and recall.

 

The Image Retrieval System is
software that provides a user in searching the images for user needs. Image
Retrieval System give the results for users with images that match their images need. Image
Retrieval System extracts the keywords form the HTML documents and assign the
weights for each keyword. The rules that a Image Retrieval System should follow
to an effective as given  follows. The Image
Retrieval System must be able to build the indexes in a reasonable amount of
time to ensure the index efficiency. The Query efficiency must be ensured to
find out whether the queries are running fast. Query Effectiveness also affects
the Image Retrieval System. A Genetic
Algorithm is a search technique used in computing to find true or
approximate solutions to optimization and search problems. Genetic algorithms
are categorized as global search heuristics. Genetic algorithms are a
particular class of evolutionary algorithms that use techniques inspired by
evolutionary biology such as inheritance, mutation, selection, and
recombination. This paper illustrates the proposed architecture of image
retrieval using genetic algorithm.

 

Many researchers proposed so many
techniques have built image retrieval systems. One such technique is
query-by-example (QBE), in which users provide visual examples of the contents
features such as the color, texture etc. However, such low-level content based
retrieval schemes have some limitations that it is not-extract and is unable to
support retrieval based on abstract concepts. Since most of the users wish to
search in terms of semantic concept rather than low level features. This situation
can be handled by using keywords along with the most relevant textual
information of images to retrieve the relevant image.

 

So it is need to purpose a
technique to extract related information from associated web pages of images.
Many techniques have been proposed on the tags of the web document such image
title, page title, link structure etc. The main problem of this technique is
lower precision of retrieval. This paper proposes a faster image retrieval
system using web crawler and genetic algorithm. The content of web pages is
divided into text and images and HTML tags. From the text, the keywords are
captured. These keywords are considered to be associated keyword to define the
meaning of the images contained in the same Web document. The keywords are used
to have built index for images. So here we are using genetic algorithm for build
index for combination keywords. The keywords from the strength matrix are
inputs to the genetic algorithm and produce output as best combination terms.

 

1.      
RELATED WORKS

Historically the researchers  proposed with different technique that built an image retrieval
system. shen et al 1 presented a
chain related terms and used more information from the web documents. The
proposed technique combines the keywords with the low-level features. The
assumption made in this method is that some of the images in the database have
been already annotated in terms of keywords. The annotation is based on the
surrounding text or speech recognition or manual annotations.

Huamin
Feng et al 2 presented a
Bootstrapping framework to annotate www images using a pre-defined set of
concepts accurately and completely through the textual and visual evidences. It
can be done by the training samples. A co-training approach that fuses
evidences from image contents and theirs associated HTML text. It based on the
supervised learning concept.

Deng Cai et al
3 presented a Hierarchical
clustering of WWW image search results using visual, textual and link
information. Initially vision based segmentation algorithm is designed to
segment a webpage into blocks. From the block containing image, the textual and
link information images are extracted. For each image, three kinds of
representation can be derived visual feature based representation, textual feature
based representation and link based representation. This approach give low level of precision and recall, the retrieval
performance is found to be lower.

This paper proposes a technique for capturing
the semantic keywords for images in associated Web documents based on the
frequency occurrence of keywords and other information pertaining to an image
and built an index for combination terms for fast indexing.

 

PROPOSED IMAGE
RETRIEVAL SYSTEM

STRENGTH
MATIX

Let H be the number of HTML page, I be the number of
images and K be the set of keywords.

H = (h1,h2,h3,??..hn),
I = (i1,i2,i3,???.it) and K = (k1,k2,k3??..km)

Where n, t and m denotes the total number of
HTML pages, images and keywords respectively. Suppose a single HTML page ?hp?
may contain only ?kq? of keywords. Now, the relation between each
keyword ?kj? where (j=1,2,3,??q) with a single HTML document can be written as

 

 

 

j=1,2 ?.q

 

The
above equation denote the association between each keyword kj in a
single HTML document ?hp’

 

KEYWORD

FREQUENCY OCCURRENCE

NORMALIZED FREQUENCY OCCURRENCE

K1

1

0.033

K2

30

1

K3

20

0.66

K4

8

0.26

K5

5

0.16

 

 

Table1.
Normalized frequency occurrence.

 

In
the above table, we can consider that not all the keywords are important. It is neccessary to consider only a set
of keywords such as the
normalized frequency occurrence of these keywords is greater than a threshold.
In our approach, we have fixed this threshold as 0.25 of the maximum normalized
frequency occurrence.

Now
it is important to estimate the strength of the keywords with the image. We
have used anchor tag, head tag, title tag, image tag as high level textual
feature. For estimating the strength value as given below

 

stg(kj?Ihp)
= Nfreq(kj?hp) + S(A tag, kj) + S(Head tag, kj)
+ S(Title tag, kj) + S(Image tag, kj)

 

In
above equation S is a matching function with either 0 or 1 as the output and
j=1,2,…q. The output value of each component of above equation consider for
associating the image with keyword.

Keyword

FOC

Nfreq(kj?hp)

S(A tag, kj)

S(Title tag, kj)

S(Head tag, kj)

S(Image tag, kj)

Strength
value

K1

1

0.033

1

0

0

0

2.033

K2

30

1

0

0

1

1

3

K3

20

0.66

1

0

0

1

2.66

K4

8

0.26

1

1

1

0

3.26

Table2. Strength matrix

In
above example the keywords which are extracted from web document. While
extracting the keywords, the stemming and stop words operations are done.  After strength matrix is calculated. The
keywords with their associated strength values give to the genetic algorithm for
generating the combination terms.

GENERATION OF COMBINATION TERMS

The
proposed architecture of the Image Retrieval System by using strength matrix
and Genetic Algorithm. There are three main components that have to take care
while designing genetic algorithm. The first code is coding the problem
solutions, next is to find a fitness function that can optimize the performance
and finally, the set of parameters including the population size, population
structure and genetic operators.

The
keywords extracted from the document collection are stored in the database. A
strength value is associated with each keyword. For making search process more
efficient, the concept of combining the keywords in the term list is
introduced. Combination of the keywords  plays an important role in retrieving the
relevant images. Here we used genetic algorithm approach to obtain the set of
the best combination of the keywords. These keywords are used to create a best
set of the term combination based on the fitness function. Thus the obtained
best combination terms are stored in a combination list. The advantages of the
proposed approach save time and retrieves the most relevant document when a
query is given.

 

Representation
of Chromosomes:

                The
strength value of each keyword is stored in the database. The sum of all the
strength value is found in the database. The mean has to be calculated and the
value has to be kept as the threshold value and then keywords are grouped
accordingly. The keywords in the term list are grouped as high strength terms
and low strength terms and stored as hstgterm list and lstgterm list
respectively.

Web
pages are scanned by the crawler and the keyword and associated frequency value
and strength value is stored. The sum of all the keyword strength value stored
in the database is found. The strength value above the mean value are termed as
high strength words and the strength value that are less than the mean value
are termed as low strength words.  Each
gene in the chromosome shows the index of the terms in the list. Let the terms
list be (1. Sachin, 2. Samsung, 3. Calendar, 4. Mobile, 5. search engine ).

Fitness
function:

                Where n is the number of times
the keywords are appearing in the whole document and N is the total number N of
documents present in the document collection.

For the chromosomes
shown above, the fitness values are given below in the table:

Chromosome
Fitness Function:

CHROMOSOME

FITNESS
FUNCTION

6,5

1.509

13,8

1.582

9,15

1.023

5,10

2.239

6,9

2.853

 

Table
5. Fitness Calculation

In
the fitness function, we can find the first two combination terms are having
more fitness than the third one. So when selection is applied we can ignores
the another chromosomes for putting the matting pool.

Genetic
Operators:

                Random selection
is used as a selection operator. The crossover used is single point crossover.
Single point mutation is used, if after generating the new population, the
fitness function is no longer improving then terminates the run.

 

The
input to the system is a set of index terms. The output obtained in the set of best combination terms
and they represent the possible solutions to the problem. The chromosomes are
randomly generated from the hstglist and the lstglist. Each chromosome is
evaluated by a fitness function. This best set of the combination terms is
applied in image retrieval system for obtaining the relevant results. Evaluate
the image retrieval system with a standard test collection using the parameters
precision and recall. Precision is the fraction of the images retrieved that
are relevant to the user?s image need. Recall is the fraction of the images
that are relevant to the query that is successfully retrieved.

In
proposed approach, user gives a query and it is searched against the image
database which as the combination terms .The combination of terms are obtained
from using genetic approach. The query is compared against the images and a
similarity measure is calculated to find out whether that particular image is
relevant to the query or not. If the image is relevant, it is retrieved. After
retrieving the relevant images from the database, sort those images and rank
them.

CONCLUSION

The
textual keywords for capturing high level semantics of an image in web
documents. The keywords present in HTML documents can be effectively used for
describing the high-level semantics of the images present in the same HTML
document. The web crawler was developed to download the web document along with
the images from World Wide Web.  Keywords
are extracted from the HTML documents after removing stop words and performing
stemming operation. The strength of each keyword is calculated and associated
with images in HTML documents. Each keyword and its corresponding strength
value is given to genetic algorithm to obtain the set of best combination of
terms. This is combination terms is used to retrieve more relevant results.
This has been verified using the evaluation measures, precision and recall. The
advantages of the proposed approach save time and retrieves the most relevant
document when a query is given.

 

REFERENCE

F.
Long, H.J. Zhang & D.D. Feng (2003) ?Fundamentals of content-based
image retrieval? Multimedia Information Retrieval and Management,
Springer, Berlin.H.
Feng, R. Shi, & T.-S. Chua (2004) ?A bootstrapping framework for
annotating and retrieving WWW images? In: Proceedings of the ACM International
Conference on Multimedia.D.
Cai, X. He, Z. Li, W.-Y. Ma & J.-R. Wen (2004) ?Hierarchical
clustering of WWW image search results using visual, textual and link
information?, In: Proceedings of the ACM International Conference on
Multimedia.Zhao,R.  & Grosky, W. I (2002) ?Narrowing the
Semantic Gap?Improved Text-Based Web Document Retrieval using Visual
Features?, IEEE Transactions on Multimedia, Vol. 4, No. (2), pp. 189-200.Jorng-Tzong
Horng & Ching-Chang Yeh (2000) ?Applying genetic algorithms to query
optimization in document retrieval?, pp 737-759.H.
Feng ; T.-S. Chua. (2003) ?A bootstrapping approach to annotating
large image collection?. Workshop on ?Multimedia Information Retrieval?,
organized in part of ACM Multimedia 2003. Berkeley, 55-62.Google
image search engine, http://images.google.com. S. N. Sivanandam and S. N. Deepa
?Introduction to Genetic algorithms?.P.Sumathy
“Ranking
images in Web Documents based on HTML TAGs for image retrieval from WWW”,
International Journal of Computational Intelligence Studies, Inderscience,
Vol.3, No.2/3, pp.176-195, 2014P.Shanmugavadivu, P.Sumathy,
A.Vadivel (2011) ?Capturing High-Level Semantics of Images in Web
Documents Using Strength Matrix?.

 

Arshi
Khan. (2006) “Content Based Image Retrieval using Genetic
Algorithm?, International
Journal of Engineering Science and Computing.Zhong Su, Hongjiang Zhang, Stan Li, and Shaoping Ma,
?Relevance Feedback in Content-Based image Retrieval: Bayesian Framework,
Feature subspaces, and Progressive Learning?, IEEE Transactions on Image
Processing, vol. 12, no. 8, August 2003.

 

Weiguo Fan, Praveen Patha and Mi Zhou. “Genetic-based
approaches in ranking function discovery and optimization in information
retrieval ? A framework?, Decision Support Systems 47 (2009) 398?407.

 

Zhengyu Zhu, Xinghuan Chen, Qingsheng Zhu, Qihong Xie ?A GA-based query
optimization method for web information retrieval?, Applied Mathematics and Computation 185 (2007) 919?930.