IRJET-Feature based Image Retrieval based on Color

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 05 Issue: 06 | June-2018 www.irjet.net p-ISSN: 2395-0072
© 2018, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 1826
Feature Based Image retrieval based on Color
Purbani Kar1, Lalita Kumari2
1,2Assistant Professor, Dept. of Computer Science & engineering, NIT Agartala, Tripura, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Image retrieval is the field of study concerned
with searching and retrieving digital images from a
collection of database. Content-based image retrieval plays
a central role in the application area such as multimedia
database systems in recent years. The emergence of
multimedia technology and the rapidly expanding image
and video collections on the Internet have attracted
significant research efforts in providing tools for effective
retrieval and management of visual data. The fundamental
idea of this approach is to generate automatically image
descriptions directly from the image content by analyzing
the content of the images. Feature extraction is the basis of
content-based image retrieval. This involves extraction of
the image features at a distinguishable extent. For color
based image retrieval, color features are the most important
elements enabling human to recognize images. For
categorizing images color features can provide powerful
information and they are used for image retrieval, so color
based image retrieval is mostly used method. Color features
of the images are generally represented by color histograms.
Before using color histograms, however, we need to select
and quantify a color space model and choose a distance
metric.
Key Words:Content based image retrieval,Color
histogram, Image features, Color space model, Distance
metric
1. INTRODUCTION
With the rapid development of image capture technology
and internet, the number of digital image is getting larger
and larger. Traditional keyword-base retrieval method
cannot work efficiently anymore. It is a process that
organized and stored the information according to a
certain way, and in accordance with the needs of users to
find the interrelated information, it is also called
Information Storage and Retrieval.The main method of
image files is to establish Keywords or text description of
the title as well as some additional information, and then
establish a link between storage path and the keywords of
the image, which is text-based image retrieval.
However, with the storage capacity of images to start
using GB or TB, the own shortcomings of text-based image
retrieval technology led to two difficulties in the retrieval:
First, it has been impossible to note each image; second,
the subjectivity and nonprecision of image annotation may
lead to the adaptation in the retrieval process.
In order to overcome these problems, a Great progress has
been achieved in the field of Content Based Image
Retrieval(CBIR) in recent years. The color feature, texture
feature, shape feature and affine invariant features have
all been used in image retrieval. Of all image content
features, color and texture are two important features and
play an important role in image content. Color histogram
has the advantages of transform invariant, rotate invariant
and scale invariant and has been widely used in image
retrieval. In This paper Three dimension color space HSV
is used and by using distance metrics a no of images
similar to the query images are retrieved. [5]
2. RELATED WORK
Content Based Image Retrieval using Texture, Color and
Shape for Image Analysis by Amanbir Sandhu and Aarti
Kochhar is a paper which describes content based image
retrieval based on features like color, texture and shape.
For Texture feature extraxtion model they have
considered Gray Level Cooccurrence matrix (GLCM). [1]
Petteri Kerminen and Moncef Gabbouj has developed a
image retrieval technique based on color matching.They
have discussed three color spaces RGB, L*a*b* and
HSV/I.They have mentioned the advantages and
disadvantages of these three color spaces.And also they
mentioned the color quantization and color histogram.In
their system the RGB components are analyzed pixel by
pixel and their real combination is the most important
thing (i.e. if a point has the values 150, 10,150 it is light
violet).[2]
In Color Based Image Retrieval System by Pawandeep
Kaur1, Sakshi Thakral2, Mandeep Singh3 they have
implemented a system quite similar to us.They have been
implemented on the image database and one query image
is chosen for getting images having almost same
histogram.They used the HSV color model, Minkowski-
form distance metrics etc in their system. They have
mentioned four requirements of color based image
retrieval system. Such as Technique to obtain the
metadata, having primitive features of images, Users query
demands evaluated by interfaces used, methods to
compare the similar or different images, Efficient indexing
and metadata storage techniques.[3]
Avneet Kaur, V. K. Banga mentioned Color moments, color
histograms, color coherence vector, color correlogram for
their system.Color moments are the statistical moments of
the probability distributions of colors. Color moments
used especially when image contain just the objects. [4]
The first order (mean), the second (variance) and the third
order (skewness) color moments have been proved to be

effective and efficent in representing color distribution of
images.
Research of Image Retrieval Based on Color by Bai Xue, Liu
Wanjun discussed two methods of feature extraction Color
histogram and Color Moment.In view of the defects of the
two extraction method of color feature color histogram
and color moment method, they select an integrated
approach of the two methods to extract the color feature
in this paper in order to improve the retrieval accuracy
and reform the ranking. [5]
3. TECHNOLOGIES USED
Color features of the images are generally represented by
color histograms. Before using color histograms, however,
we need to select and quantify a color space model and
choose a distance metric.
3.1 Color Model:
There are many color models to express color such as the
RGB color model, YUV color model and the HSV color
model. In this system we have used the HSV color space.
The computer can only identify the RGB color component
of an image, in which R represents the red component, G
represents the green component, B represents the blue
component[5].
Therefore, we need the following formula for the image
conversion from RGB color space to HSV color
3.2 Three-Dimensional Color Histogram:
3.2.1 Quantization of HSV Color Vectors:
A (16:4:4) non-uniform quantization method is adopted in
which H vector is divided into 16 values and S, V are
divided into 4 values separately. The S, V and H values
after quantization are shown below.
3.2.2 The Construction of Three-Dimensional Color
Histogram:
In the image after quantization, its H vector has 16 values
and S, V has 4 values separately. So this paper defines an
array of 16*4*4 size which is T to calculate the ratio of
pixels of each color to the overall pixels. The elements in T
is defined in
Where T(i, j, k) means the ratio of pixels whose color value
are the ith value in H, the jth value in S and the kth value in V
to the overall pixels of the image.Ni;j;k is the number of
pixels whose color values are values mentioned above. M
is the overall pixel number of the image.
3.2.3 Similarity measures:
Define sample image as I and the image to be matched is J
and their three dimensional color histograms are FI and FJ.
Histogram intersection is performed to determine
similarity of these two images.
Where M is the overall number of pixels in the image,min
() is a function that can give the smallest value. The value
range of S (I,J) is 0 to 1. The more similar the two images
are, the bigger the value of S is. For the same images, S is 1.
Many similarity measures have been developed for image
retrieval based on empirical estimates of the distribution
of features in recent years. We denote D (I, J) as the
distance measure between the query image I and the

image J in the database; and fi (I) as the number of pixels in
bin i of I.
3.2.4 Minkowski-Form Distance:
Minkowski-form distance is the most widely used metric
for image retrieval. Minkowski-form distance metrics
compare only the same bins between color histograms. .If
each dimension of image feature vector is independent of
each other and is of equal importance, the Minkowski-
form distance Lp is appropriate.
When p=1, 2, and, D (I, J) is the L1, L2 (also called Euclidean
distance), and L distance respectively. The Histogram
intersection can be taken as a special case of L1 distance,
which is used by Swain and Ballard to compute the
similarity between color images.
Where A= [aij] is a similarity matrix, and aij denotes the
similarity between bin Fi and Fj. and are vectors that list all
the entries in and Quadratic form distance can lead to
perceptually more desirable results than Euclidean
distance.
3.2.5 User Interaction
Two main features of user interaction in image retrieval
system are as follows:
3.2.6 Query Specification
Various types of queries can be listed as simple visual
feature query, feature combination query, localized
feature query, query by example, user-defined attribute
query, object relationship query, and concept queries.
User can have two ways to make distinction. In first way
user looks for category search and in second method he
searches for target search.
3.2.7 Relevance Feedback
The iterative and automatic refinement of a query is
known as relevance feedback in information retrieval
literature.
Relevance feedback can be seen as a form of supervised
learning to adjust the subsequent queries using the
information gathered from the user’s feedback.
Fig-1: Flow of relevance feedback
4. STEPS OF THE RETRIEVAL SYSTEM
4.1Training
Almost 700 images have been used for populating the
database. For each image a 3-D histogram of its HSV values
is computed. At the end of the training stage, all 3D HSV
histograms are stored in the same .mat file.
4.2Query
In order to retrieve M (user-defined) query results, the
following steps are executed:
1. The 3D (HSV) histogram of the query image is
computed. Then, the number of bins in each direction (i.e.,
HSV space) is duplicated by means of interpolation.
2. for each image i in the database:
 Load its histogram Hist (i).
 Use interpolation for duplicating the number of
bins in each direction.
 For each 3-D hist bin, compute the distance (D)
between the hist of the query image and the i-th
database image.
 Keep only distances (D2) for which, the respective
hist bins of the query image are larger than a
predefined threshold T (let L2 the number of these
distances).
 Use a 2nd threshold: find the distance (D3) values
which are smaller than T2, and let L3 be the
number of such values.
 The similarity measure is defined as: S (i) = L2 *
average (D3) / (L32).
3. Sort the similarity vector and prompt the user with the
images that have the M smaller S values.

Fig-2: Image retrieval system based on color
5. RESULTS AND DISCUSSION
The methodology discussed in the earlier chapter has been
implemented on the image database and one query image
is chosen for getting images having almost same
histogram. HSV histogram is used for comparison.
Here steps of implementation have been shown in 5.1 and
5.2.The experiments are performed under computer of
Intel Core i5 CPU, 2.30Hz, 4G RAM and with operating
system of Windows 7.
The size of images is 250*225. For an image set of 606
images the average time is 4 second which is very fast.
Fig-3:Images Get Compared And Their Equivalent Hsv
Histograms
Fig-4: Eleven most resembling images with the query
image
6. CONCLUSION AND FUTURE SCOPE
Content Based Image Retrieval is now became a burning
topic in computer science. It is getting more and more
importance to practical applications.This paper introduces
an effective image retrieval method which is based on the
color feature.
Color is usually represented by the color histogram, color
correlogram, color coherence vector, and color moment
under a certain color space. Here, in this paper 3D-HSV
color histogram has been used.Up to now; the Minkowski
and Quadratic form distance are the most commonly used
distances for image retrieval. To set up an indexing
scheme, dimension reduction is usually performed to
reduce the dimensionality of the visual feature vector.
Query results can be refined through the relevance
feedback of users.
Although color-based image retrieval provides an
intelligent and automatic solution the majority of current
techniques are based on low level features. In general,
each of these low level features tends to capture only one
aspect of an image property. Neither a single feature nor a
combination of multiple features has explicit semantic
meaning. Although relevance feedback provides a way of
filling the gap between semantic searching and low-level
data processing, this problem remains unsolved and more
research is required.
REFERENCES
[1] Content Based Image Retrieval using Texture, Color
and Shape for Image Analysis,Amanbir Sandhu ,Rayat
Bahra College of Engg & Nanotechnology for
Women,Hoshiarpur (Pb), Aarti Kochhar, Council for
Innovative Research International Journal of
Computers & Technology www.ijctonline.com ISSN:
2277-3061 Volume 3, No. 1, AUG, 2012
[2] IMAGE RETRIEVAL BASED ON COLOR MATCHING,
Petteri Kerminen, Moncef Gabbouj, Pori School of
Technology and economica, TUT(Tampere University
of Technology), Information Technology Signal

Processing Laboratory P.P Box 30, P.O Box 553 28601
Pori 33101 Tampere.
[3] Color Based Image Retrieval System Pawandeep Kaur,
Sakshi Thakral, Mandeep Singh,IOSR Journal of
Computer Engineering (IOSRJCE) ISSN: 2278-0661
Volume 1, Issue 5 (May-June 2012), PP 01-05
www.iosrjournals.org
[4] Color Based Image Retrieval Avneet Kaur, V. K. Banga
[5] Research of Image Retrieval Based on Color Bai Xue,
Liu Wanjun School of Software, Liaoning Technical
University, Hulidao, 125105, China, 2009
International Forum on Computer Science-Technology
and Applications xiaopanghuamao@163.com
[6] Color Image Retrieval Based on Color and Texture
Features, Xiuxin Chen, Kebin Jia, Xiaoqin Lian, and
Shiang Wei , D. Jin and S. Lin (Eds.): Advances in CSIE,
Vol. 1, AISC 168, pp. 669674. Springerlink.com
Springer-Verlag Berlin Heidelberg 2012
[7] Efficient CBIR Using Color Histogram Processing,
Neetu Sharma, Paresh Rawat and jaikaran Singh
,Signal & Image Processing : An International
Journal(SIPIJ) Vol.2, No.1, March 2011
[8] Color-based retrieval Nicu Sebe *, Michael
S.Lew,Leiden Institute of Advanced Computer Science,
Department of Computer Science, Niels Bohrweg 1,
Leiden, CA 2333, The Netherlands Received 21
February 2000; received in revised form 14 July 2000
[9] Dr. Fuhui Long, Dr. Hongjiang Zhang, Prof. David Dagan
Feng, Fundamentals of contentbased Image retrieval,
Chapter 1.
[10] Swain MJ and Balland DH. Color indexing of Computer
Vision [J], 1991-7(1):11-32.

IRJET-Feature based Image Retrieval based on Color

Recommended

More Related Content

What's hot (17)

Similar to IRJET-Feature based Image Retrieval based on Color (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET-Feature based Image Retrieval based on Color