ISSN:2229-6093
P V N Reddy et al, Int. J. Comp. Tech. Appl., Vol 2 (4), 1016-1020
Color and Texture Features for Content Based Image Retrieval
P. V. N. Reddy
K. Satya Prasad
Research Scholar, Dept. of ECE,
Jawaharlal Nehru Technological University
Kakinada, Andhra Pradesh, India.
pvnreddy_alfa@rediffmail.com
Rector &Professor Dept. of ECE,
Jawaharlal Nehru Technological University
Kakinada, Andhra Pradesh, India.
Prasad_kodati@yahoo.co.in
Abstract
Content based image retrieval (CBIR) has been one of
the most important research areas in computer science
for the last decade. A retrieval method which combines
color and texture feature is proposed in this paper.
According to the characteristic of the image texture, we
can represent the information of texture by Multi
Wavelet transform. We choose the color correlogram in
RGB color space as the color feature. The experimental
t results show that this method is more efficient than the
traditional CBIR method based on the single visual
feature and other methods combining color and texture.
2. Color
Color feature is one of the most widely used features in
low level feature [6]. Compared with shape feature and
texture feature, color feature shows better stability and
is more insensitive to the rotation and zoom of image.
Color not only adds beauty to objects but also more
information [1], which is used as powerful tool in
content-based image retrieval. In color indexing, given
a query image, the goal is to retrieve all the images
whose color and texture compositions are similar to
those of query image. In color image retrieval there are
various methods, but here we will discuss some
prominent methods.
1. Introduction
Application of World Wide Web (www) and the
internet is increasing exponentially, and with it the
amount of digital image data accessible to the users. A
huge amount of Image databases are added every
minute and so is the need for effective and efficient
image retrieval systems. There are many features of
content-based image retrieval but four of them are
considered to be the main features. They are color,
texture, shape, and spatial properties. Spatial properties,
however, are implicitly taken into account so the main
features to investigate are color, texture and shape.
Though there are many techniques of search this paper
will focus on color and texture features for CBIR. The
main motivation of the present work is to use the Multi
Wavelet decomposition scheme and color correlogram,
which yield improved retrieval performance. Through
combination of Multi wavelet decomposition and color
correlogram we can increase the number of features,
which in turn improves the retrieval accuracy. To
support the efficient and fast retrieval of similar images
from image databases feature extraction plays an
important role in content-based image retrieval. A
fundamental ingredient for content based image
retrieval is the technique used for comparing images.
IJCTA | JULY-AUGUST 2011
Available online@www.ijcta.com
Typical characterization of color composition is
done by color histograms [7]. In 1991 Swain and
Ballard [2] proposed the method, called color indexing,
which identifies the object using color histogram
indexing. Color histograms are way to represent the
distribution of colors in images where each histogram
bin represents a color in a suitable color space (RGB
etc) [3]. A distance between query image histogram and
a data image histogram can be used to define similarity
match between the two distributions. To overcome
problem with histogram in 1995 Mehtre et al [4]
proposed two new color- matching methods as
“Distance Method” and “Reference Color Table
Method”, for image retrieval. They used a coarse
comparison of the color histograms of the query and
model images in the Distance method they proposed.
Most color histograms are very sparse and thus
sensitive to noise. In 1995 Stricker and Orengo [5]
proposed cumulated color histogram. Their results are
better than color histogram approach. Observing the
fact that the color histograms lack information about
how color is spatially distributed, in 1997 Rui and
Huang [6], introduced a new color feature for image
1016
ISSN:2229-6093
P V N Reddy et al, Int. J. Comp. Tech. Appl., Vol 2 (4), 1016-1020
retrieval called color correlogram. This feature
characterized how the spatial correlation of pairs of
color changes with distance in an image. Usually,
because the size of color correlogram is quite large, the
color auto corrlogram is often used instead. This feature
only captures spatial correlation between identical
colors. The main contributions of this paper are as
follows. Here we have proposed a Multi wavelet-based
approach is used for texture feature extraction and color
correlograms are used for color feature extraction for
CBIR. These color and texture features are combined
to improve the retrieval efficiency.
Multiwavelets were defined using several wavelets
with several scaling functions [10]. Multiwavelets have
several advantages in comparison with scalar wavelet
[8]. The features such as compact support,
Orthogonality,
symmetry,
and
high
order
approximation are the base features for this transform.
A scalar wavelet cannot possess all these properties at
the same time. On the other hand, a Multiwavelet
system
can
simultaneously
provide
perfect
representation while preserving length (Orthogonality),
good performance at the boundaries (via linear-phase
symmetry), and a high order of approximation
(vanishing moments) [9]. Thus Multiwavelets offer the
possibility of superior performance and high degree of
freedom for image processing applications, compared
with scalar wavelets. The study of Multiwavelets was
initiated by Goodman, Lee and Tang. The special case
of Multiwavelets with multiplicity 2 and support (0, 2),
was studied by Chui and Lian. When a multi resolution
analysis is generated using multiple scaling functions
and wavelet functions, it gives rise to the notion of
Multiwavelets [10]. A Multiwavelet with ‘r’ scaling
functions and ‘r’ wavelet functions is said to have
multiplicity ‘r’. When r = 1, with one scaling function
and one wavelet function, the Multiwavelet system
reduces to scalar wavelet system. In Multiwavelet
transforms they have two or more scaling functions and
wavelet functions. The set of scaling functions are
represented using the vector notation
(1)
Where
is called the multi-scaling function. The
Multiwavelet function is defined from the set of
wavelet function
IJCTA | JULY-AUGUST 2011
Available online@www.ijcta.com
wavelets. Multiwavelets differ from scalar wavelet
systems in requiring two or more input streams to the
Multiwavelet filter bank. Multiwavelets are an
extension of the scalar wavelet to the vector case. As in
the scalar wavelet case, the theory of Multiwavelets is
based on the idea of multi resolution analysis (MRA).
The difference is that Multiwavelets have several
scaling functions. The multi scaling function and the
Multiwavelet function will satisfy matrix dilation
equations,
φ (t ) = 2 ∑ H k φ (2t − k )
∞
3. Multi Wavelet Transform
φ (t ) = [φ1 (t )φ 2 (t )..........φ r (t )]T
(2)
ψ (t ) = [ψ 1 (t )ψ 2 (t )...........ψ r (t )]T
When r = 1, ψ (t ) is called a scalar wavelet or simply
(3)
k = −∞
∞
ψ (t ) = 2 ∑ Gk φ (2t − k )
(4)
k = −∞
The filter coefficients Hk and Gk are N by N matrices
instead of scalar. Corresponding to each Multiwavelet
system, there is a matrix-valued with multi-rate filter
bank. A Multiwavelet filter bank has “taps” that are N
× N matrices. One desirable feature of any transform
used in image retrieval is the amount of energy
compaction achieved. A filter with good energy
compaction Properties can decorrelate a fairly uniform
input signal into a small number of scaling coefficients
containing most of the energy and a large number of
sparse wavelet coefficients. Therefore better
performance is obtained when the wavelet coefficients
have values clustered about zero with little variance.
Thus Multiwavelets have the potential to offer better
representative quality than the conventional scalar
transforms. Finally, Multiwavelets can achieve better
level of performance than scalar wavelets with similar
computational complexity. Wavelets are useful tools
for image processing applications such as image
retrieval and denoising.
LL
LH
HL
HH
L1L1
L1L2
L1H1
L1H2
L2L1
L2L2
L2H1
L2H2
H1L1
H1L2
H1H1
H1H2
H2L1
H2L2
H2L1
H2L2
Figure1. Image decomposition after a single
level decomposing for (a) Scalar wavelets and
(b) Multi-wavelets.
1017
ISSN:2229-6093
P V N Reddy et al, Int. J. Comp. Tech. Appl., Vol 2 (4), 1016-1020
4. Proposed Algorithm
The basic steps involved in the proposed CBIR
system includes database processing and resizing,
creation and normalization of feature database,
comparison and image retrieval. Steps of the proposed
algorithm are as follows.
A. Texture feature extraction:
1.
2.
Figure2. Conventional iteration of Multiwavelet
decomposition.
During a single level of decomposition using a
scalar wavelet transform, the 2- D image data is
replaced by four blocks corresponding to the sub bands
representing either low pass or high pass in both
dimensions. These sub bands are illustrated in Figure.
1. The Multi-wavelets used here have two channels, so
there will be two sets of scaling coefficients and two
sets of wavelet coefficients. Since multiple iteration
over the low pass data is desired, the scaling
coefficients for the two channels are stored together.
Likewise, the wavelet coefficients for the two channels
are also stored together. The Multi-wavelet
decomposition sub bands are shown in Figure.2. For
Multi-wavelets the L and H have subscripts denoting
the channel to which the data corresponds. For
example, the sub band labeled L1H2 corresponds to
data from the second channel high pass filter in the
horizontal direction and the first channel low pass filter
in the vertical direction. This shows how a single level
of decomposition is done. In practice, there is more
than one decomposition performed on the image.
Successive iterations are performed on the low pass
coefficients from the previous stage to further reduce
the number of low pass coefficients. Since the low pass
coefficients contain most of the original image energy,
this iteration process yields better energy compaction.
After a certain number of iterations, the benefits gained
in energy compaction becomes rather negligible
compared to the extra computational effort. Usually
five levels of decomposition are used.
The main title (on the first page) should begin 1-3/8
inches (3.49 cm) from the top edge of the page,
centered, and in Times 14-point, boldface type.
Capitalize the first letter of nouns, pronouns, verbs,
adjectives, and adverbs; do not capitalize articles,
coordinate conjunctions, or prepositions (unless the
title begins with such a word). Leave two 12-point
blank lines after the title.
IJCTA | JULY-AUGUST 2011
Available online@www.ijcta.com
3.
Convert all data base images into gray images.
Decompose each image in the Multi wavelet
domain.
Compute the standard deviation σ k on each
4.
sub band of the Multi Wavelet decomposed
image.
The resulting SD vector is
f = [σ 1 , σ 2 , σ 3 ,........σ n ]
B. Color feature extraction:
1.
2.
3.
4.
5.
Load the image.
Separate the R, G, and B spaces from the
image.
Quantize the each color space into 32 levels.
Apply the correlogram in 00, 450, 900, and
1350 on each color space.
Construct the feature vector by using
correlogram.
C. Combined feature
Form the combined feature vector by
concatenating the color feature and texture feature
D. Apply query image and calculate the combined
feature vector as given in steps A to B.
E. Calculate the similarity using Euclidean distance.
F.
Retrieve all relevant images to query image based
on minimum “Euclidean distance”.
5. Experimental Results
Comparison of average retrieval accuracy for 640
different colored textures using correlogram and Multi
Wavelet transform is provided. When Tile 10.bmp is
given as query and retrieved using three different
methods of retrieval used, in this paper following
results were obtained. The results obtained by both the
1018
ISSN:2229-6093
P V N Reddy et al, Int. J. Comp. Tech. Appl., Vol 2 (4), 1016-1020
texture features and color features consideration the
retrieval efficiency is 82% that is fourteen images
retrieved from the database are of same texture and
color contents. Red, Green and blue curves in figure 1
indicates the average retrieval efficiency using only
color, only texture and color and texture combined
respectively.
texture is 512x512. Each 512x512 image is divided into
sixteen 128x128 non-overlapping sub-images, thus
creating a database of 640 (40x16) images. The
performance of the proposed method is measured in
terms of average retrieval rate (ARR) is given by Eq.
(5).
The MIT VisTex database used in our experiment
consists of 40 different textures [11]. The size of each
ARR =
1
DB
∑ R ( I , n)
DB
i =1
i
(5)
n =16
Number of top matches considered
Method
1
2
3
4
5
6
7
8
9
10
MWT
100
96.7
94.3
91.8
89.8
87.9
86.3
84.5
82.9
81.3
CC
100
99.5
99.2
98.4
97.6
96.5
95.5
94.5
93.2
91.8
MWT+CC with L1 distance
100
99.5
99.1
98.2
97.2
96.2
94.9
93.5
92.1
90.5
MWT+CC with Euclidean
100
99.2
97.6
96.2
94.2
92.8
91.1
89.5
87.9
85.8
MWT+CC with d1 distance
100
99.6
99.2
98.5
97.7
96.7
95.8
94.6
93.6
92.6
6. Conclusions
After the absolute analysis of the results obtained by
each method following conclusions can be drawn.
When only color is considered as retrieval parameter in
CBIR gives only 62.5% of average retrieval efficiency.
Similarly when only texture features are considered as
retrieval parameter there is not much improvement in
the retrieval efficiency. The average retrieval efficiency
obtained by this method is only 68.75%. This shows
that only texture features or only color features are not
sufficient to describe an image. But there is
considerable increase in retrieval efficiency when both
color and texture features are combined for CBIR. The
average percent retrieval efficiency has increased up to
75%. Thus it is rightly said in [1] that only color or
only texture cannot differentiate a cheetah and a tiger.
7. References
[1] Manesh Kokare, B.N. Chatterji and P.K. Biswas,
"A survey on current content based image retrieval
methods", IETE Journal of Research, Vol. 48, No.3
and 4, May-Aug 2002.
[2] Swain, M.J., and Ballard, D.H. "Color indexing".
Int’l Journal of Computer Vision, 1991, Vol.7(1),
1132.
[3] R.C. Gonzalez, R.C. Woods, Digital Image
Processing, Addison-Wesley, Reading, MA, 1992.
[4] Babu M Mehtre, M S Kankanhalli, A Desai
Narasimhalu, and Guo Chang Man, “Colour
matching for image retrieval”,P R L, 16, pp 325331, 1995.
[5] Stricker and M Orengo, “Similarity of Colour
Images, Proc SPIE Storage and Retrieval for image
and video Databases”, 1995.
IJCTA | JULY-AUGUST 2011
Available online@www.ijcta.com
1019
ISSN:2229-6093
P V N Reddy et al, Int. J. Comp. Tech. Appl., Vol 2 (4), 1016-1020
[6] Th.Gevers (2001). “Color Based Image Retriev-al” .
Springer Verlag GmbH. pp.886-917
[7] M.J. Swain, D.H. Ballard (1991). “Color indexing”
. Int. J. Comput. Vis. 7 11-32.
[8] Hiremath.P.S, Shivashankar . S. Wavelet based
features for texture classification, GVIP Journal,
Vol.6, Issue 3, pp 55-58, December 2006.
and symmetry properties”. Proc. IEEE Conf. On
Image Processing , vol. II, pp, 1429.
[19] N. G. Kingsbury (2001). “Complex wavelets for
shift invariant analysis and filtering of signals”
.Journal of Applied and Computational Harmonic
Analysis, vol. 10, no. 3, pp. 234-253.
[20] N. Kingsbury (1998). “The dual tree complex
[9] X. Wan and C. C. J. Kuo. Color distribution
wavelet transform: A new efficient tool for image
analysis and quantization for image retrieval. In
restoration and enhancement”. Proc of EUSIPCO
SPIE proceedings, vol.2670, February 1996.
[10] Michael B. Martin and Amy E. Bell , “New Image
Compression Techniques using Multiwavelets and
98, pp. 319-322.
[21] Subrahman yam Murala(2009) . “Color and Texture
Features for Image Indexing and Retrieval”.
Multiwavelet Packets”, IEEE Transactions on
image processing, vol. 10, No. 4, , April 2001.
[11] MIT Vision and Modeling Group, Vision Texture.
Available: http://vismod.www.media.mit.edu .
[12] B.S. Manjunath and W.Y. Ma (1996). “Texture
P.V.N. Reddy, Research scholar
in ECE Department University
College of Engineering, JNTUK,
Kakinada. A.P.
features for browsing and retrieval of image
data”,IEEE Trans. Pattern Anal. Mach. Intell, vol.
Email:pvnreddy_alfa@rediffmail.com
8, no. 8, pp. 837-842.
[13] M. N. Do and M. Vetterli (2002). “Wavelet-based
texture retrieval using generalized Guassian density
and Kullback-leibler distance”. IEEE Trans.Image
Process, vol. 11, no. 2, pp. 146-158.
[14] Challa S. Sastry *, M. Ravindranath (2007). “A
modified Gabor function for content based image
retrieval”, Pattern Recognition Letters. pp, 293-300.
[15] Ju Han , Kai-Kuang Ma (2007). “Rotation-invariant
and scale-invariant Gabor features for texture image
retrieval”. Image and Vision Computing. Vol. 12.
Dr. K. Satya Prasad is currently
working as Rector, JNTUK
Kakinada
&
Professor
of
Electronics & Communications
Engineering. He has more than
28 years of Experience in
teaching and 20 years of R & D. He is an expert in Digital
Signal Processing. He has produced 4 PhD’s and guiding 10
PhD scholars. He has published more than 30 technical
papers in national and International Journals and
conferences.
pp,1474-1481.
[16] Manesh Kokare *, P.K. Biswas (2007). “Texture
Email:Prasad_kodati@yahoo.co.in
image retrieval using rotated wavelet filters”.
Pattern Recognition Letters Vol.28. pp, 1240-1249.
[17] N. G. Kingsbury (1999). “Image processing with
complex wavelets,” Philos.Trans. R. Soc. London A
Math. Phys. Sci, vol. 357, no. 3, pp. 2543-2560.
[18] N. G. Kingsbury (1994). “A Dual-Tree Complex
Wavelet Transform with improved orthogonality
IJCTA | JULY-AUGUST 2011
Available online@www.ijcta.com
1020