Ug 205-image-retrieval-using-re-ranking-algorithm-11

ISSN:2348 9510
International Journal Of Core Engineering & Management(IJCEM)
Volume 1, Issue 1, April 2014
49
Image Retrieval using Re-Ranking Algorithms
Mayuri D. Joshi , Revati M. Deshmukh, Kalashree N.Hemke, Ashwini Bhake
Computer Technology Department , Yeshwantrao Chavan College Of Engineering, Nagpur,
Maharashtra India.
mayuri1992joshi@gmail.com,rewtideshmukh@gmail.com,kalashreehemke@gmail.com,
bhake.ashwini@gmail.com.
ABSTRACT
Our objective is to improve the performance of
keyword based image search engines by re-
ranking their original results. Present search
engines face some limitations. First is that,
many irrelevant results are obtained along
with the demanded images. Secondly,
duplication of images is normally seen in the
image database retrieved from any search
engine for the entered query word .Third is
that , user has to surf along long pages for
expected result or set of result images, leading
to confusion many a times.
Re-ranking of images is a solution to
these limitations. The objective of this work is
to reduce user efforts and semi-automatically
generate a large number of images for a
specified object class which are more accurate
as compared to the normally retrieved images.
Candidate images are obtained by a text based
web search querying on the object identifier
(e.g., the word penguin). The WebPages and
the images they contain are downloaded. The
task is then to remove irrelevant images and
re-rank the remainder. So the main motive of
the proposed system is to complement the task
of present search engines for the interest of
user . This system is a combination of web and
desktop application that is active on user or
client side. This work will re-rank the images
as well as form clusters of similar images and
also notify the user to which cluster the
entered query image will belong .
1. Introduction
Image retrieval is one of the major issues of
user concern. The most common way of image
retrieval is the text based image retrieval
technique (TBIR) [6]. The working of TBIR
needs to be understood. Suppose user enter a
query in the search box of Google, how exactly
Google returns the result?
When we enter a keyword in the search box,
Google navigates through different pages
looking for the occurrence of that word. Each
image has a tagging associated with it. Google
looks for the occurrence of the entered keyword

ISSN:2348 9510
50
in these tagging. As Google returns the result
based on the search of text, it is text based
image retrieval.
.
Figure 1. Working of Google search engine. [7]
After retrieving the images, Google then
performs top-rank sequencing. In this, Google
ranks the images on the basis of number of hits.
The images are ranked in descending order of
number of hits i.e. the image with maximum
number of hits is ranked first. Thus Google
performs text- based image retrieval for
retrieving the images and top-rank sequencing
to rank the retrieved images and then returns the
result.
This technique of TBIR is not sufficient for
accuracy hence the trend moves towards content
based image retrieval (CBIR). Our work aims at
extracting colour features of images and using
histograms to re-rank the images to improve the
performance.
2. Literature Survey
Figure 2. Architecture of image harvesting and
re-ranking system [5]
Content based image retrieval uses content of
image for retrieval. The architecture diagram
(Fig. 2) [5] gives an overview of Content Based
Image Retrieval (CBIR). Content of image
refers to the colour, shape and texture of the
image.

ISSN:2348 9510
51
This system is based on the following
functionalities and features:
a) Extraction
Extraction means deciding the feature based on
which re-ranking will be performed and
extracting the same. If the entered query is
‘sunset’ colour will be the appropriate feature
for re-ranking. But if we need to differentiate
between ‘cotton’ and ‘snow’ texture will be the
appropriate feature for discrimination.
b) Distance calculation and similarity
measurement:
This step calculates the difference between the
images in terms of corresponding chosen
feature. The images are plotted in feature space
and then the distance between them is
calculated using any of the following formulae.
Given two feature vectors A and B such that
A= a1 B= b1
a2... b2...
Euclidean distance is given by:
City block is another approach for distance
measurement. [3]
Figure 3. Distance calculation and measurement
Lesser the distance more similar the images
will be. As mentioned in [5], for CBIR
implementation, image classification should be
fast and efficient. CBIR emphasises on use of
visual content of image like colour, texture,
shape etc. for image comparison and retrieval
rather than textual query. We have considered
colour feature for comparison.
c) Following the distance calculation and
similarity measurement, Re-ranking is
performed. Various algorithms can be used to
perform re-ranking. Our implementation uses
K-mean and hierarchical algorithm.
3. Implementation and Experimental
results
Details of Completed Modules with supporting
results
The completed modules are
1. Web Module
2. Desktop Module

ISSN:2348 9510
52
1. The Web Module
This module is used for downloading the
images from search engine on the client
machine where further processing will be done.
This module downloads bulk of images for a
given query on a single click. For experimental
purpose, we have downloaded 100 images for
keyword ‘Rose’.
Figure 4. Part of dataset used
2. The Desktop Module
The downloaded images are further processed in
this module. There are two sub-modules-
I. Indexing Module
This module performs clustering by using K-
means and hierarchical clustering algorithms in
combination.
K-means Algorithm
 Input
 k: the number of clusters
 D: a dataset containing n
elements
 Output: a set of k clusters
 Method
(1) arbitrarily choose k
elements from D as the initial
cluster mean values
(2) repeat
(3) assign each element
to the cluster whose mean the
element is
closest to
(4) once all of the
elements are assigned to
clusters, calculate the actual
cluster
means
(5) until there is no change
between the new and old
cluster means
Hierarchical clustering Algorithm
This algorithm is used in bottom up
fashion for comparing images and
assigning the clusters based on
threshold and computed distances.
The working can be understood from
the figure

ISSN:2348 9510
53
Figure 5. Hierarchical clustering
The indexing module uses these two
algorithms to cluster the images based on the
color feature of image. Following steps are
followed-
1. The Red, Green and Blue content
of an image are extracted and
plotted as histograms and
histogram matrices are being
generated.
2. The indexing module takes the
number of clusters k as input.
3. It then applies K-means for the
iterations and hierarchical
clustering for image classification.
4. The distance calculation in K-
means is done by using the
histogram matrices
5. The indexing modules returns
number of clusters containing re-
ranked images.
Figure 6. Indexing module home page. Shows
indexing in progress
For experimental purpose we have obtained
result for four clusters. They are as follows.
Figure 7. Cluster1 of 4 obtained after indexing.

ISSN:2348 9510
54
Figure 8. Cluster2 of 4 obtained after indexing.
Figure 9. Cluster 3 of 4 obtained after indexing.
Figure 10. Cluster 4 of 4 obtained after
indexing.
II. Search Module
This module is used to find the relevant
cluster of the given input image. It takes an
image as input and returns a cluster containing
relevant images based on distance matching
obtained in indexing module.
Figure 10. Result of search module for input
image.
4. Applications
This project finds application in various fields
like medical sciences for disease diagnosis , in
astronomy , mechanical engineering and other
fields where image clustering and detection is
required.
5. Future Scope
 Different features like texture, edges
can be used for Re-Ranking.
 An intelligent Re-Ranking System
Capable of identifying which feature
should be used and Re-rank
accordingly
6. Conclusion
Basic thing reviewed from this paper is that the
text-based image retrieval is not sufficient for
obtaining precise images for a given query.
Thus techniques based on CBIR are found to be
more vibrant and are likely to be adopted for
such applications. This implemented work is

ISSN:2348 9510
55
efficient for image clustering and classification
with considerable accuracy.
The domain of image harvesting, retrieval and
re-ranking offers a vast scope for exploration as
well as innovation and has a broad future scope.
7. References
[1] Venkat N.Gudivada, Vijay V. Raghavan
"Content-Based Image Retrieval Systems" IEEE
Transaction 0018-9162, 1995 .
[2] Edward Remias, Gholamhosein
Sheikholeslami, Aidong Zhang." Block-
Oriented Image Decomposition and Retrieval in
Image Database Systems". IEEE Transaction 0-
8186-7469-5, 1996.
[3]Szabolcs Sergy´an, Budapest Tech, John von
Neumann ,Faculty of Informatics." Color
Histogram Features Based Image Classification
in Content-Based Image Retrieval Systems".6th
International IEEE Symposium on Applied
Machine Intelligence and Informatics-2008.
[4] Yihun Alemu, Jong-bin Koh, Muhammed
Ikram, Dong-Kyoo Kim." Image Retrieval in
Multimedia Databases: A Survey". Fifth
International Conference on Intelligent
Information Hiding and Multimedia Signal
Processing ,IEEE-2009.
[5] K.A. Shaheer Abubacker, L.K. Indumathi."
Attribute Associated Image Retrieval and
Similarity Re-ranking". Proceedings of the
International Conference on Communication
and Computational Intelligence – 2010, Kongu
Engineering College, Perundurai, Erode,
T.N.,India.27 – 29 December,2010.pp.235-240.
[6] Lixin Duan, Wen Li, Ivor Wai-Hung Tsang,
and Dong Xu, Member, IEEE. "Improving Web
Image Search by Bag-Based Re-ranking".IEEE
TRANSACTIONS ON IMAGE PROCESSING,
VOL. 20, NO. 11, NOVEMBER 2011.
[7] J. Sivic and A. Zisserman. Video Google:
A text retrieval approach to object matching in
videos. In Proc. ICCV, 2003.

Ug 205-image-retrieval-using-re-ranking-algorithm-11

More Related Content

What's hot (19)

Viewers also liked (10)

Similar to Ug 205-image-retrieval-using-re-ranking-algorithm-11 (20)

More from Ijcem Journal (20)

Recently uploaded (20)

Ug 205-image-retrieval-using-re-ranking-algorithm-11