MICANS INFOTECH
www.micansinfotech.com
9003628940
2014 International Conference on Electronic Systems, Signal Processing and Computing Technologies
Enhancement of image retrieval by using colour,
texture and shape features.
Ms. Apurva N. Ganar
Prof. C. S. Gode
Dept. of Electronics and telecom.Engg. Dept. of Electronics and telecom.Engg.
Y.C.C.E, Nagpur, Maharashtra (India) Y.C.C.E, Nagpur, Maharashtra (India)
[email protected]
[email protected]
Abstract—Content based image retrieval technique is done by three
primitive methods namely through color, shape and texture. This
paper provides specified path to use these primitive features to
retrieve the desired image. The technique by which we obtain the
required image is CBIR. In CBIR first the HSV color space is
quantified to obtain the color histogram and texture features. Using
these components a feature matrix is formed. Then this matrix is
mapped with the characteristic of global color histogram and local
color histogram, which are analysed and compared. For the cooccurrence matrix between the local image and the images in the
database to retrieve the image. For extracting shape feature gradient
method is used here. Based on this principle, CBIR system uses
color, texture and shape fused features to retrieve desired image from
the large database and hence provides more efficiency or
enhancement in image retrieval than the single feature retrieval
system which means better image retrieval results.
Keywords— Global color histogram, local color histogram, RGB,
HSV, co-occurrence matrix, gradient method.
Prof. Sachin M. Jambhulkar
Dept. of Electronics and Engg.
R.G.C.E.R Nagpur, Maharashtra (India)
[email protected]
manually enter keywords for images in a large database can be
inefficient, expensive and may not capture every keyword that
describes the image. Thus a system that can filter images based on
their content would provide better indexing and return more accurate
results.
There is a growing interest in CBIR because of the
limitations inherent in metadata-based systems, as well as the large
range of possible uses for efficient image retrieval. Textual
information about images can be easily searched using existing
technology, but this requires humans to manually describe each
image in the database. This is impractical for very large databases or
for
images
that
are
generated
automatically,
e.g.
those
from surveillance cameras. It is also possible to miss images that use
I. INTRODUCTION
different synonyms in their descriptions. Systems based on
The recent tremendous growth in computer technology has also
brought a substantial increase in the storage of digital imagery.
Examples of applications can be found in everyday life, from
museums for archiving images or manuscripts, to medicine where
millions of images are generated by radiologists every year. Storage
of such image data is relatively straightforward, but accessing and
searching image databases is intrinsically harder than their textual
counterparts. The goal of Content-Based Image Retrieval (CBIR)
systems is to operate on collections of images and, in response to
visual queries, extract relevant image. The application potential of
CBIR for fast and effective image retrieval is enormous, expanding
the use of computer technology to a management tool.The more
realistic approach taken in the early 1990s was to work with simple
low level features instead such as the colour histograms used by
Swain and Ballard. Since then many more sophisticated methods
have been developed. However, due to the difficulties involve most
practical approaches are still rooted in low level feature extraction
and description.
categorizing images in semantic classes like "cat" as a subclass of
"animal" avoid this problem but still face the same scaling issues.
II. RELATED WORK
Most of the search engines (ex.google, yahoo, etc.,) are
based on a semantic search, i.e., the user types in a series of
keywords and the images are also annotated using keywords. Thus
the match is done primarily through these keywords. In the recent
years CBIR system have been developed to handle the large image
database effectively. Basically color, texture and shape have been
used for extracting similar images from an image database. Different
CBIR techniques have adopted different techniques. Some
techniques have used global color and texture features (3,4,5) where
as some have used local colors and texture features (6,7,8). After that
the method of segmentation is proposed, where the image is
segmented into regions based on color and texture features. And then
region to region similarity is done. The regions are closed to human
perception and are used as basic building blocks of computation of
feature and similarity measurement. These systems are known as
region based image retrieval. But image segmentation algorithm is
difficult in the human perception point of view. To ensure robustness
against such inaccurate segmentation region maintaining algorithm is
used. Color space is divided into small ronges. Each interval is
regarded as a bin. Then matching techniques are carried out (1).
Shape is another important feature for perceptual object
recognition. Various applications proposed techniques including the
chain codes, characteristic, circumference, area and circular
degree(9). N Grace Selvarni and Dr. S. Annuradai (12) used genric
The "Content-based" means that the search analyzes
the contents of the image rather than the metadata such as
keywords, tags, or descriptions associated with the image. The
term "content" in this context might refer to colors, shapes,
textures, or any other information that can be derived from the
image itself. CBIR is desirable because most web-based
image search engines rely purely on metadata and this
produces a lot of garbage in the results. Also having humans
978-1-4799-2102-7/14 $31.00 © 2014 IEEE
DOI 10.1109/ICESC.2014.48
251
MICANS INFOTECH
www.micansinfotech.com
9003628940
fourier shape descriptor technique. High rate of computation is
observed here. Combination of edge histogram and fouries
transforms for computing edge image is proposed in literature (11).
Gradient method for shape feature extraction and retrieval of similar
image from image database is reported in literature (2). The literature
(2) proposed the gradient method for extraction of shape features
from the image and to retrieve the similar image from the
database.This paper uses color, texture and shape information for
image retrieval to enhance the image retrieval results to a better
efficiency.
III. System overview and proposed methods
1. Colour feature extraction:
Color is the first and most straightforward visual feature for
Indexing and retrieval of images, relatively robust and simple to
represent. It is also the most commonly used feature in the field.
Color has been an active area of research in image retrieval, more
than in any other branch of computer vision. The choice of a color
system is of great importance for the purpose of proper image
retrieval. An important criterion is that the color system is
independent of the underlying imaging device. This is required when
images in the image database are recorded by different imaging
devices such as scanners, camera's and cam recorder (e.g. images on
Internet). Another Prerequisite might be that the color system should
exhibit perceptual uniformity meaning that numerical distances
within the color space can be related to human perceptual differences.
This is important when images are to be retrieved which should be
visually similar (e.g. stamps, trademarks and paintings).
Colour histogram is a representation of the distribution of
colors in an image. Colour histogram represents the image but from
another perspective. It counts similar pixels and store it. Basically,
colour histogram is a colour descriptor and each descriptor
contains a feature extraction algorithm and matching function.
1.1 Local versus global colour histogram based CBIR:
Colour histogram is divided into global colour histogram
and local colour histogram. Here, we are using both the local and
global colour histogram and then we are comparing their results. As
our one of the aim is to show how the enhancement of image
retrieval can be done by using local colour histogram instead of
global colour histogram.
Fig.1.1.1Flowchart for Global colour histogram based CBIR
Next step is to count each feature value. Then by using
Euclidean distance formula shown below, calculate similarity.
1.1.1Global colour histogram based CBIR:
It is colour histogram is the most known colour histogram
used to detect similar images.
RGB colour space does not meet the visual requirements. So the
image converted from RGB to HSV space.HSV space is used as it is
very common and MATLAB command is easily available for RGB
to HSV conversion.
A and B are two feature vectors. n is the dimension of A and
B.
The flowchart shows steps for image retrieval by using global colour
histogram. Firstly RGB colour space is converted to the HSV colour
space. H stands for hue, which varies from 0 to 360 degrees. S stands
for saturation. It shows the grey range in colour space and ranges
from 0 to 1. V stands for value. It is the brightness of colour and
varies with saturation. It ranges from 0 to 1.
1.1.2 Local colour histogram based CBIR:
In a Local colour histogram based CBIR, the image is
divided into NXN tiles. The size of the tile should not be too large or
too small. Here, the size of the tile considered as 3x3, as it is found to
be more effective. Now for each block we repeat the same steps as
explained in global colour histogram based CBIR.By comparing the
results of Global colour histogram based CBIR and Local colour
histogram based CBIR, it can be observed that local colour histogram
The next step is to quantify the images using the following
formula:
252
MICANS INFOTECH
www.micansinfotech.com
9003628940
based CBIR is more accurate in image retrieval. Global colour
histogram based CBIR calculates frequency of colour. So, the spatial
distribution of colour information is lost. So, the Local colour
histogram based CBIR is found to be more effective than Global
colour histogram based CBIR.
2. Texture feature Extraction:
Texture is a repetitive tone in an image. For
calculating texture feature, co-occurrence matrix are to be
computed. The flow chart shows the steps for extracting the
texture feature from the images.
K, M and L,n are changes of selected calculating windows, # is the
pixel logarithm.
The four texture parameters capacity, entropy, moment of inertia and
relevance are calculated. Mean and standard deviation of each
parameter is taken as each component of texture features.
Internal normalization is the last step. For an image Ii, feature vector
is Hi=[hi,1,hi,2,……………hi,N]. the Gaussian normalization concept
is used to calculate internal normalization to make each feature of the
same weight.
mj- mean
ơj- standard deviation
3. Shape feature extraction:
We are using gradient method to extract shape feature.
An image is a function of two variables f(x,y). if image is assigned
value from 0 to 1 according to brightness of the imag then for white
colour, pixel is assigned a zero value. For black colour of pixel value
1 is assigned and for greyish colour, value between 0 and 1 is
assigned depending on the brightness of that pixel. Rapid change in
colour intensity i.e. the sharp contrast indicates the edge in an image.
A rapid change in a function gives a large magnitude of the gradient
at edges. The gradient is the geometric computing method for
characterizing symmetric breaking of an ensemble of asymmetric
vectors regularly distributed in a square lattice. In gradient method,
edges are detected first. It is done by looking for the maximum and
minimum in the first derivative of the image.
Fig.2.1 Texture feature Extraction
RGB colour space is converted to grey scale, as we are considering
only texture feature here. So the colour feature can be neglected.
Also, the computation of grey scale image is simpler than coloured
image. The conversion from RGB to grey scale is one by using the
formula shown below.
Y=0.29xR+0.58xG+0.114xB
Y- Grey scale value
R,G, B- Red, Green and Blue colour
Then the grey scale quantification is done. As the grey scale is 256,
the respective co- occurrence matrix computed is 256x256. So the
grey scale of the image is compressed to reduce the computational
efforts. 16 compression levels are chosen here to improve the speed
of extraction of texture feature. Next step is feature value calculation.
Four co-occurrence matrices are found using the formula shown
below.
Mean measures the average grey level in an image
characteristics in x and y direction, about each pixel. Where,
x is the horizontal direction and y is the vertical direction.
Mean is computed as:
Mean(k1)=(f(x))1/2
253
MICANS INFOTECH
www.micansinfotech.com
9003628940
Mean(k2)=(f(y))1/2
The minima are calculated by using the following formula.
Minima1=1.5626 Mean(k1)
Minima2=1.5626 Mean(k2)
The minima gives the measure of average contrast. And
1.5626 is a constant empirically determined. Compare the
absolute value of the f(x) with the minima 1 store the strong
edge on x direction and compare the absolute value of f(y)
with the minima 2 store the strong edge on y direction.
Masking x and y directions to obtain sharp edge of the image.
IV.Simulation Results
Fig.c: CBIR using Texture feature
Fig.a: input image for retrieval
Fig. d: CBIR using Color and Texture fused features
The figure a shows the query image to be retrieved. The figure b
shows the retrieved images by using color histogram .Figure c shows
the results of CBIR using texture feature. And figure d shows the
CBIR using Color and Texture fused features. We can see the
difference between CBIR using color and texture feature and the
CBIR using color and texture fused features. In CBIR using color
and texture fused features, images retrieved are the best matching
images to the query image. So, from the results we can say that
CBIR using color and texture features is more robust. The work of
this literature is completed up to comparison between CBIR using
color and texture and CBIR based on color and texture fused features.
The future work will be on image retrieval using color, texture and
shape fused features.
Fig.b: CBIR using Color feature
V. CONCLUSION
On the basis of previous researches, the paper explored
low-level features of color and texture extraction of CBIR. After
comparing the CBIR based on color and texture features with that of
the color and texture fused features, it is observed that CBIR based
on color and texture fused features provides better results i.e. results
of color and texture fused features are robust than the color and
texture features based image retrieval.
Other low level feature such as shape will be fused to make the
image retrieval more efficient in future.
254
MICANS INFOTECH
www.micansinfotech.com
VI. REFERENCES
[1] Jun Yue, Zhenbo Li, Lu Liu, Zetian Fu. “Content based image
retrieval using color and texture fused features”, 1121-1127, 2011,
Elsevier.
[2] Jagadeesh Pujari, Pushpalata S. N., “Content based image
retrieval using color and shape descriptors”,2010 IEEE.
[3] W. Niblack et al., “The QBIC project: Querying images by
content using color, texture and shape” in Proc.SPIE, vol. 1908, San
Jose, CA, pp. 173-187,Feb. 1993.
[4] A. Pentland, R. Picard, and S. Sclaroff, “Photobook: Contentbased manipulation of image databases”, in Proc. SPIE storage and
retrieval for image and video databases II, San Jose, CA, pp. 34-47,
Feb. 1994.
[5] M. Stricker, and M. Orengo, “Similarity of color images”, in
Proc. SPIE storage and retrieval for image and video databases, pp.
381-392, Feb. 1995.
[6] C. Carson, S. Belongie, H. Greenspan, and J. Malik, “Blobworld:
Image segmentation using expectation- maximization and its
application to image querying.” In IEEE trans. On PAMI, Vol. 24,
No. 8, pp. 1026-1038,2002.
[7] A. Nastev, R. Rastogi, and K. Shim, “WALRUS: A similarity
retrieval algorithm for image databases”, in proc. ACM SIGMOD
Int. conf. management of data, pp. 395-406, 1999.
[8]J. Li, J. Z. Wang, and G. Wiederhold, “IRM: integraed region
matching for image retrieval”, in proc. Of the 8th ACM int. conf. on
multimedia, pp. 147-156, oct. 2000.
[9] Yong-xianga sun, cheng-minga zhang, pingzenga liu, hong-mei
zhu, “The shape feature extraction based on chain code”, proceedings
of the 2007 inter. Conf. on wavelet analysis and pattern recognition,
Bejing, china, 2-4 Nov.,2007.
[10] Sami Brandit, Jorma Laaksoner and krkki oja, “Statistical shape
features in content- based image retrieval”, 2000 IEEE.
[11]A. Grace Selvarni and Dr. Annadurai, A “Content based medical
image retrieval for medical images using generic fourier descriptors”,
Journal of computational intelligence in bioinformatics, ISSN: 0973385X vol. 1 number 1 (2008) pp. 65-72, Research Indian
Publications.
255
9003628940