Efficient Approach for Content Based Image Retrieval Using Multiple SVM in YACBIR

Natarajan Meghanathan et al. (Eds) : ACSIT, SIPM, CMIT, CoNeCo - 2016
pp. 19–29, 2016. © CS & IT-CSCP 2016 DOI : 10.5121/csit.2016.60803
EFFICIENT APPROACH FOR CONTENT
BASED IMAGE RETRIEVAL USING
MULTIPLE SVM IN YACBIR
Lakhdar LAIB1
and Samy Ait-Aoudia2
1,2
National High School of Computer Science ESI, Algiers, ALGERIA
l_laib@esi.dz
s_ait_aoudia@esi.dz
ABSTRACT
Due to the enormous increase in image database sizes, the need for an image search and
indexing tool is crucial. Content-based image retrieval systems (CBIR) have become very
popular for browsing, searching and retrieving images in different fields including web based
searching, industry inspection, satellite images, medical diagnosis images, etc. The challenge,
however, is in designing a system that returns a set of relevant images i.e. if the query image
represents a horse then the first images returned from a large image dataset must return horse
images as first responses. In this paper, we have combined YACBIR [7], a CBIR that relies on
color, texture and points of interest and Multiple Support Vector Machines Ensemble to reduce
the existing gap between high-level semantic and low-level descriptors and enhance the
performance of retrieval by minimize the empirical classification error and maximize the
geometric margin classifiers. The experimental results show that the method proposed reaches
high recall and precision.
KEYWORDS
Content Based Image Retrieval, YACBIR, Feature extraction, Multiple Support Vector
Machines, Classification
1. INTRODUCTION
With fast and exceptional growth in digital technology, a significant increase in the number of
images produced by scientific, educational, medical, industrial, virtual museums, individual
photograph collections and other applications raises several practical issues.
Content-based Image Retrieval (CBIR) system [1] [2] [3] [4] is the application of computer vision
techniques which is used to extract similar images from an large image database. CBIR system is
mainly used in various applications such as art collections, education and training, crime
prevention, military field, architectural and engineering design, journalism and advertising, and
web searches [5].
Simply stated, a CBIR system retrieves images corresponding to a query image by examining the
image contents (low-level feature). Many image low-level features such as color, texture and
shape are widely used [7]. The color aspect can be treated by techniques like averaging and
histograms. The texture aspect can be achieved by using four descriptors: contrast, entropy,
energy and inverse differential moment [8]. The shape aspect can be extracted by gradient or
morphological operators [6] or by considering points of interest. [8].

20 Computer Science & Information Technology (CS & IT)
Figure 1: Content Based Image Retrieval Approach.
The main goal in CBIR system is searching the similar images from the database based on their
content. To accomplish the retrieval task, CBIR system firstly computes and stores the feature
vector for each image in the database. When the query image is introduced in the system, its
feature vector will be extracted and compared to those of other images in the database. Then a
series of similar images is returned to the user. Different similarity measures are used in the
matching process [6].
This paper is organized as follows. We give in section 2 an overview of our proposed system. In
section 3, concepts of SVM (Support Vector Machine) are reminded. Experimental results on
sample datasets are given in section 4. Section 5 gives conclusions.
2. OVERVIEW OF PROPOSED SYSTEM
2.1. Method summary
Step 1: Feature extraction f (Color, Texture, and Point of interest).
Step 2: Sum of weighted color, texture and points of interest features.
Step 3: Query image feature classification by Multiple Support Vector Machines (Find the
class Label).
Step 4: Compute the distance between the Query image and images in the dataset (Class
Label).
Step 5: Output Retrieved images (Similar images).

Computer Science & Information Technology (CS & IT) 21
Figure 2: Overview of the process
2.2. Feature extraction in YACBIR
Feature extraction is a common way to improve the performance of the visual features and
improve the efficiency of the CBIR systems. The best subset of features is selected by a heuristic
search algorithm to improve the precision of CBIR system. Three types of features are used for
image indexing [7].
Color - HSV
Texture -contrast, entropy, energy and inverse differential moment
Point of interest -Harris detector
The YACBIR system combines three characteristics of an image to compute a weighted similarity
measure. The characteristics of the image are the color, texture and points of interest. Color and
texture characteristics used are global while points of interest are local shape characteristics.
The similarity measure used in YACBIR is a sum of weighted color, texture and points of interest
(shape) similarity measures. This similarity measure is given by:
S = α.Sc+ β.St+ γ.Ss with (α+β+γ) = 1

2.3. Support Vector Machines (SVM)
In machine learning, support vector machines (SVM) are supervised learning models with
learning algorithms. Given a set of data where each data is marked as belonging to one of two
categories, a SVM training algorithm constructs a model which classifies new data in one of these
two classes. Though SVMs are fundamentally developed for such binary classification case, they
are extendable for multi-class problems [9].
The SVM algorithm consists of two phases:
a. Training phase:
Input learning data for a SVM classifier can consist of distance to border vectors, binary images,
Zernike moments, and more. Each input data is represented by vector ix with label
liiy ≤≤±= 1,1 , l is the number of samples.
The decision boundary should classify all points correctly, thus
( ) ib
iiy ∀≥+ ,1xTw .
The decision boundary can be found by solving the following constrained optimization
problem:
( ) ib
ii
ytosubjectMinimize ∀≥+ 1
2
2
1
xTww
The Lagrangian of this optimization problem is:
i
i
i
b
ii
y
i
L ∀≥∑ 




 −



 +−= 01
2
2
1
αα xTww T
the optimization problem can be rewritten in terms of
i
α by setting the derivative of the
Lagrangian to zero:
( )
i
i
i
y
ii
tosubject
ji
j
T
ij
y
i
y
ji
i
i
WMaximize
∀∑ =≥
∑−∑=
0,0
,2
1
αα
αααα xx

Figure 3: Support Vector Machine (SVM).
This quadratic programming problem is solved when:
( )( ) i
ib
ii
yi xxTw ∀=−+ 01α with 0>
i
α
b. Testing phase:
The resulting classifier is applied to unlabelled images to decide whether they belong to the
positive or the negative category. The label of x is simply obtained by computing:
∑
=
=
s
j j
t
j
t
y
j
t
1
xw α with ( )sjjt ,,1 Κ= the indices of the s support vectors. Classify z
as category 1 if the sum is positive, and category 2 otherwise.
3. EXPERIMENTAL RESULTS
The experiments in this section are performed on a workstation with an Intel Corei3 2.10 GHz
CPU and 4 GB of RAM running Win 10.
We evaluate the effectiveness of the proposed method for Corel-1k which is a set of 1000 images
that are subdivided into 10 sufficiently distinct semantic classes, including African people,
seaside, bus, building, mountain, dinosaur, elephant, horses, flowers and food images some
examples are depicted in Figure 4.

Figure 4: Image database used (Corel-1k).
3. 2 Precision-recall
Precision-versus-recall curve is the basic evaluation measure in information retrieval. They are
defined as follow:
ds retrievevant imageer of reletotal numb
ievedmages retrrelevant inumber of
=Precision ………… [19]
atabases in the dvant imageer of reletotal numb
ievedmages retrrelevant inumber of
=Recall ..….. [20]
Table 1: Comparison precision proposed method with other methods

Figure 5. Precision of proposed approach compared to other methods for different semantic classes in the
Corel database.
Table 2: Comparison recall proposed method with other methods
Class
Lin et
al.[10]
Elalami
[11]
Poursist-
aniet al.
[12]
Guo et
al.[13]
Subra-
et al
[14]
Walia
et al.
[15]
Irtaza et
al. [16]
Elalami
[17]
Zeng et
al.[18]
Proposed
Method
African 14.10 15.30 14.04 16.94 13.95 10.20 13.00 16.10 14.50 14.46
Beach 19.20 19.80 8.88 9.08 10.85 18.00 12.00 20.30 13.04 10.98
Monuments 17.40 18.20 14.16 13.56 12.79 11.60 12.40 19.10 14.12 14.79
Buses 12.10 11.60 15.26 17.06 17.93 15.60 17.00 12.60 17.84 18.07
Dinosaurs 10.10 9.80 20.00 19.86 19.74 20.00 18.60 10.90 20.00 19.08
Elephants 14.90 15.60 12.76 14.22 9.76 16.80 13.00 16.30 14.10 16.60
Flowers 11.20 11.80 18.48 18.66 18.46 20.00 18.80 12.90 18.96 19.28
Horses 13.40 13.90 18.94 19.16 17.89 20.00 15.40 14.40 18.36 18.37
Mountains 21.30 22.80 11.24 9.96 9.46 16.80 14.60 23.60 14.45 14.34
Food 13.20 13.80 14.90 16.16 14.18 7.60 16.20 14.80 15.76 16.73
Average 14.68 15.20 14.86 15.46 14.50 15.66 15.10 16.10 16.11 16.27

Figure 6. Precision of Proposed Approach Compared to Other Methods for Different Semantic Classes in
the Corel Database

4. CONCLUSION
In this paper, we proposed an efficient learning approach based on multiple SVMs (Support
Vector Machines) and YACBIR system. From the test results, we show that retrieval performance
of the proposed algorithm has quite good result in the image retrieval system. In future research,
other classification techniques can be tested. Artificial Neural Networks (ANN) transfer functions
and multiple SVMs kernel can be used to improve the classifier performance.
REFERENCES
[1] Query by image and video content: The QBIC system” S. W. Teng and G. Lu, “Codebook Generation
in Vector Quantization Used for Image Retrieval,” International Symposium on Intelligent
Multimedia and Distance Education, 1999.
[2] Y. Liu, D. Zhang, G. Lu, and W. Ma, “A Survey of Content-based Image Retrieval with High-level
Semantics,” Pattern Recognition, Vol. 40, No. 1, 2007, pp. 262-282.

[3] Datta, Ritendra, et al. "Image retrieval: Ideas, influences, and trends of the new age." ACM
Computing Surveys (CSUR) 40.2 (2008): 5.
[4] Lew, Michael S., et al. "Content-based multimedia information retrieval: State of the art and
challenges." ACM Transactions on Multimedia Computing, Communications, and Applications
(TOMM) 2.1 (2006): 1-19.
[5] Ms. K.Arthi, Mr. J. Vijayaraghavan. (2013)”Content Based Image Retrieval Algorithm Using Colour
Models”, Chennai.
[6] Meenakshi Madhugunki, Dr. D.S.Bormane. (2011)”Comparsion of Different CBIR Techniques”,
Pune.
[7] Ait-Aoudia, Samy, Ramdane Mahiou, and Billel Benzaid. "YACBIR: yet another content based
image retrieval system." Information Visualisation (IV), 2010 14th International Conference. IEEE,
2010. Esmat
[8] Rashedi, Esmat, and Hossein Nezamabadi-pour. "Improving the precision of CBIR systems by feature
selection using binary gravitational search algorithm." Artificial Intelligence and Signal Processing
(AISP), 2012 16th CSI International Symposium on. IEEE, 2012.
[9] M. Mr.A.H.Karode and M. W.Pawade, “Texture image classification using support vector machine.”
International Journel of Computing Technology and Applications, 2012.
[10] C.H. Lin, R.T. Chen, Y.K. Chan, A smart content-based image retrieval system based on color and
texture feature, Image Vis. Comput. 27 (6) (2009) 658–665.
[11] M.E. ElAlami, A novel image retrieval model based on the most relevant features, Knowl.-Based
Syst. 24 (2011) 23–32.
[12] P. Poursistani, H. Nezamabadi-pour, R.A. Moghadam, M. Saeed, Image indexing and retrieval in
JPEG compressed domain based on vector quantization, Math.Comput. Model. 57 (5–6) (2013) 1005–
1017.
[13] J.M. Guo, H. Prasetyo, H.S. Su, Image indexing using the color and bit pattern feature fusion, J. Vis.
Commun. Image R 24 (2013) 1360–1379.
[14] M. Subrahmanyam, Q.M.J. Wu, R.P. Maheshwari, R. Balasubramanian, Modified color motif co-
occurrence matrix for image indexing and retrieval, Comput. M. Saeed, Image indexing and retrieval
in JPEG compressed domain based on vector quantization, Math.A. Irtaza, M.A. Jaffar, E. Aleisa,
T.S. Choi, Embedding neural networks for semantic association in content based image retrieval,
Multimed. Tool Appl. 72 (2) (2014) 1911–1931.
[15] E. Walia, A. Pal, Fusion framework for effective color image retrieval, J. Vis.Commun. Image R 25
(2014) 1335–1348.
[16] A. Irtaza, M.A. Jaffar, E. Aleisa, T.S. Choi, Embedding neural networks for semantic association in
content based image retrieval, Multimed. Tool Appl.72 (2) (2014) 1911–1931.Image retrieval using
spatiograms of colors quantized by Gaussian Mixture Models. Neurocomputing Volume 171, 1
January 2016, Pages 673–684 72 (2) (2014) 1911–1931.
[17] M.E. ElAlami, A new matching strategy for content based image retrieval system, Appl. Soft Comput.
14 (2014) 407–418.
[18] S. Zeng, R. Huang, H. Wang, Z. Kang Image retrieval using spatiograms of colors quantized by
Gaussian Mixture Models. Neurocomputing Volume 171, 1 January 2016, Pages 673–684

[19] K. Mikolajczyk and C. Schmid, Scale and affi ne invariant interest point detectors, Int. J. Comput.
Vis. 1 (2004), 63 – 86.
[20] T. P. Minka and R. W. Picard, Interactive learning using a society of models, Pattern Recogn. 30
(1997), 565 – 581.
AUTHORS
Lakhdar LAIB received the B Eng and MSc degrees in computer science from École
nationale Supérieure en Informatique (Algeria) in 2010 and 2013, respectively. Since
2014, he has been pursuing PhD studies at École nationale Supérieure en Informatique
(Algeria) under the supervision of Professor Samy Ait-Aoudia His primary research
interests include statistical models and application to image segmentation, computer
vision and pattern recognition.
Samy Ait- Aoudia received a DEA “Diplôme d'Etudes Approfondies” in image
processing from Saint-Etienne University, France, in 1990. He holds a Ph.D. degree in
computer science from the Ecole des Mines, Saint-Etienne, France, in 1994. He is
currently a Professor of computer science at the National High School in Computer
Science at Algiers/Algeria, where he is involved in teaching BSc and MSc levels in
computer science and software engineering. His areas of research include image
processing, CAD/CAM and constraints management in solid modeling.

Efficient Approach for Content Based Image Retrieval Using Multiple SVM in YACBIR

Recommended

More Related Content

What's hot (15)

Similar to Efficient Approach for Content Based Image Retrieval Using Multiple SVM in YACBIR (20)

Recently uploaded (20)

Efficient Approach for Content Based Image Retrieval Using Multiple SVM in YACBIR