Content based image retrieval (cbir) using

International Journal of Computational Science and Information Technology (IJCSITY) Vol.2, No.2, May 2014
DOI : 10.5121/ijcsity.2014.2204 33
CONTENT BASED IMAGE RETRIEVAL (CBIR) USING
MULTIPLE FEATURES FOR TEXTILE IMAGES BY
USING SVM CLASSIFIER
Mr.P.Anand 1
(AP/ECE), T.Ajitha2
, M.Priyadharshini3
and M.G.Vaishali4
Velammal college of Engineering and Technology, Madurai-09.
ABSTRACT
In this project, we proposed a Content Based Image Retrieval (CBIR) system which is used to retrieve a
relevant image from an outsized database. Textile images showed the way for the development of CBIR. It
establishes the efficient combination of color, shape and texture features. Here the textile image is given as
dataset. The images in database are loaded. The resultant image is given as input to feature extraction
technique which is transformation of input image into a set of features such as color, texture and shape.
The texture feature of an image is taken out by using Gray level co-occurrence matrix (GLCM). The color
feature of an image is obtained by HSI color space. The shape feature of an image is extorted by sobel
technique. These algorithms are used to calculate the similarity between extracted features. These features
are combined effectively so that the retrieval accuracy and recall rate is enhanced. The classification
techniques such as Support Vector Machine (SVM) are used to classify the features of a query image by
splitting the group such as color, shape and texture. Finally, the relevant images are retrieved from a large
database and hence the efficiency of an image is plotted.The software used is MATLAB 7.10 (matrix
laboratory) which is built software applications
KEYWORDS
CBIR, HSI color space, SVM, Feature extraction.
I.INTRODUCTION:
The course of textile images plays a vital role in content based image retrieval (CBIR). Thus the
task is capably storing, processing and retrieving textile image records. "Content-based" means
that the search will analyze the actual contents of the image rather than the metadata such as
keywords, tags, and/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 metadatass itself. Our proposed work is mentioned as steps given below.
Step 1: Loading image
a). Convert the image into workable format with the extension as .bmp or .jpg.
b). Convert the workable format into gray scale image.
Step 2: Feature Extraction
a). The texture, color and shape features are extracted gray level co-occurrence matrix,
HSI color space, sobel technique respectively.
b). Make the feature vector by making the average value of each pixels byte.
Step 3: Similarity computation
a). Compute the similarity on the basis of distance above and below the threshold value.
Step 4: Mark the relevant and irrelevant images.

34
Step 5: Classify the image set.
Step 6: Train the SVM classifier.
Section II details about Block Diagram. Section III describes about Feature Extraction
Techniques. Section IV explained about an Optimization techniques.
II.BLOCK DIAGRAM
II. FEATURE EXTRACTION
In this project three features are extracted by using appropriate techniques. Color features are
extracted using HSI color space, shape features are extracted using sobel technique and texture
features are extracted using gray level co-occurrence matrix. These features are explained below.
1. Color feature extraction:
Textile image is given as a query input image. For each and every image the RGB color space is
converted into HSI color space. Then the single image is partitioned into three regions namely
Hue, Saturation and Intensity. The properties of color such as mean, standard deviation and
skewness are determined only for hue color space image. These properties are described below.
A) Mean:
Mean is defined as the average value of a color in an image. The mean of an image is considered
as a first central moment. The mean value is calculated for each region by using the formula
given below.

35
Ei=∑ Pij
B) Standard deviation:
The standard deviation is defined as the square root of the variance of distribution. The standard
deviation of an image is considered as a second central moment. The standard deviation value is
calculated for each region by using the formula given below.
i=√( ∑ )
Where, A=(Pij-Ei)2
C) Skewness:
Skewness is defined as a measure of the degree of asymmetry in the distribution. The skewness of
an image is considered as a third central moment.
Si=∛( ∑ )
Where, B= (Pij-Ei) 3
The skewness value is calculated for each region by using the formula given above.
Where,
I is the current channel index (ie) H=1,S=2, V=3
Ei1, Ei2 are the first moment (mean) of two image.
i is the second moment (standard deviation) of two image distribution.
Si is the third moment (skewness) of two image distribution.
The values are obtained as follows.
Images Mean Standard
deviation
Skewness
1.jpg 0.3483 0.2467 -0.1130
2.jpg 0.4006 0.2663 -0.7362
3.jpg 0.3708 0.2730 -0.5192
4.jpg 0.4090 0.3240 -0.4316
5.jpg 0.0353 0.1461 5.3085
6.jpg 0.4191 0.3021 -0.5950
7.jpg 0.5053 0.4396 -0.2396
8.jpg 0.4111 0.2186 -0.6466
9.jpg 0.4500 0.2646 -0.9391
10.jpg 0.3712 0.2175 -0.1779
11.jpg 0.3543 0.2976 -0.2670
12.jpg 0.3950 0.2826 -0.2002
13.jpg 0.6765 0.4343 -0.8275
14.jpg 0.2978 0.2787 -0.0621
15.jpg 0.5172 0.4638 -0.1682
Table.3.1 Color Feature Values

36
2. Shape Feature Extraction
The input image is converted into gray scale image and applying Sobel technique for edge
preservation. Eccentricity and centroid are the properties focused on shape feature extraction.
These features are explained below.
A) Eccentricity:
It is the ratio of major axis to minor axis. It characterizes the length of ROI. Lateral mean and
longitudinal mean values are calculated for eccentricity. The eccentricity is obtained by using the
formula given below.
∈=(( 20- 02)2+4 112)/ ( 20+ 02)2
B) Centroid:
The point at the centre of image called centre of area. The coordinates of centroid are the average
(arithmetic mean) of the coordinates of points in image. The centroid linkage uses the Euclidean
distance between the centroid of two clusters.
d(r,s)=||Xr-Xs||2
The values are listed in table 3.2
Images Lat
mean1
Lat
mean2
Lon
mean1
Lon
mean2
1.jpg 0.5328 0.5481 0.5034 0.5340
2.jpg 0.5119 0.5025 0.5184 0.5029
3.jpg 0.5249 0.5281 0.4900 0.5179
4.jpg 0.5073 0.5201 0.4949 0.4969
5.jpg 0.5066 0.4749 0.5032 0.5005
6.jpg 0.5479 0.5268 0.5187 0.4970
7.jpg 0.5145 0.5334 0.5008 0.5109
8.jpg 0.5000 0.5305 0.5847 0.4654
9.jpg 0.5267 0.4968 0.5182 0.4817
10.jpg 0.5276 0.5191 0.5019 0.5116
11.jpg 0.4966 0.5129 0.4588 0.4654
12.jpg 0.5139 0.5331 0.4807 0.5024
13.jpg 0.5218 0.5348 0.5128 0.5054
14.jpg 0.5216 0.4998 0.4984 0.5030
15.jpg 0.5480 0.5363 0.4992 0.4967
Table.3.2 Shape Feature Values
3. Texture Feature Extraction:
The texture features are extracted by using Gray Level Co-occurrence Matrix (GLCM). It is used
to calculate how often pixels with gray level value i occurs horizontally adjacent to a pixel with a
value j. Finding co-occurrence matrix is to remove redundancy. The features extracted are
Energy, Entropy, Contrast and Correlation. These values are obtained by using formula. Matching
is done with query image and database images.

37
Fig 3.3 GLCM
A) Contrast:
It measures local variations in gray level co-occurrence matrix. It returns a measure of intensity
contrast between a pixel and its neighbor over the whole image. The formula for contrast is given
below. .
B) Correlation:
It measures joint probability occurrence of specified pixel pairs. It returns a measure of how
correlated a pixel to its neighbor of over whole image. The range lies within [-1,1]. Correlation is
1 for positively image. It’s -1 for negatively image. Nan is for constant image.
C) Energy:
It provides the sum of squared elements in GLCM known as uniformity or angular second
moments. Range lies within range [0 1].
Energy is 1 for constant image.
D) Homogeneity:
It measures closeness of distribution of elements in GLCM to GLCM diagonal. Returns a value
that measures the closeness of the distribution of elements in the GLCM to the GLCM diagonal.
Homogeneity is 1 for diagonal matrix. The range lies within [0 1].
The values are listed in table.3.4

38
Images Contrast Correlation Energy Homogeneity
1.jpg 0.1373 0.9767 0.3177 0.9677
2.jpg 0.0921 0.9768 0.4960 0.9823
3.jpg 0.1568 0.9657 0.4294 0.9708
4.jpg 0.5820 0.9666 0.4684 0.9727
5.jpg 0.1390 0.9691 0.4653 0.9785
6.jpg 1.6310 0.8339 0.2140 0.8266
7.jpg 0.3134 0.9713 0.3147 0.9534
8.jpg 0.1393 0.8427 0.3432 0.9317
9.jpg 0.4422 0.9714 0.3169 0.9454
10.jpg 0.4132 0.9619 0.3201 0.9395
11.jpg 0.3396 0.9734 0.4283 0.9712
12.jpg 0.2585 0.9166 0.3067 0.9548
13.jpg 0.4239 0.9556 0.2464 0.9131
14.jpg 0.3988 0.9603 0.3256 0.9435
15.jpg 0.4625 0.9693 0.2569 0.8940
Table.3.4 Texture Feature Values
IV. SUPPORT VECTOR MACHINE (SVM):
SVM is suitable to classify the Color and shape features accurately and efficiently. It is
superintend learning process which analyze and recognize patterns such as texture, color, shape
and it is used for classification and regression techniques. The classification is accomplished by
hyper plane which has outsized distance to nearest distance training data points of any class
(color, texture, shape). Therefore the observation is larger the margin, lower the generalization
error of classifier. The classification separates data into training samples and testing samples. The
objective of SVM is to predict the target values of test data by giving only the test data attributes.
The steps are to be followed:
1. Transform the data to the format of SVM which is suitable for further process.
2. Applying scaling on data.
3. Consider the RBF kernel function
K(x;y)= exp( - α||x-y||2 )
4. By using the best parameter C, cross validation to train the entire training set.
5. Finally testing is needed to check the error is within acceptable limit.
It is a non parametric study with binary classifier. It is efficient because it accepts more input
variables. It only depends upon selection of hyper plane and kernel parameter.
V) RESULTS AND DISCUSSION:
Initially the features of color, shape, texture for textile images extracted are tabulated. The fig.5.1
shows that the given input image is textile T-Shirt image. The input RGB image is converted into

39
hue image. The color features are extracted using HSI color space. The output is displayed as Hue
image.
Fig.5.1 Color Feature Extraction
The given input textile image is displayed as shown in fig.5.2. The sobel technique is applied to
given input image. The edge is preserved by calculating the gradients. The preserved edge is
shown below in fig.5.2.
Fig.5.2 Shape Feature Extraction
The given input image is displayed in fig.5.3. The given RGB image is converted into gray scale
image. The texture features are extracted using Gray level Co-Occurrence matrix. The output
image is displayed in fig.5.3.
Fig.5.3 Texture Feature Extraction
The three features such as color, texture, shape are extracted. The features are fed into SVM
classifier. The query image is sent and the three features are extracted similarly and fed to
classifier.
The classifier output is described below in fig.5.4.1the color feature extracted SVM output.
Fig.5.4.1. The color feature extracted SVM output

40
Fig.5.4.2.The texture feature extracted SVM output.
Fig.5.4.3.The Shape feature extracted SVM output.
Finally the similar images are retrieved. and then compare the accuracy and retrieval time in our
project.
Fig.5.4 compared accuracy for proposed work and existing work
0
20
40
60
80
100
120
140
160
180
Retrieval time
(secs*10^2)
Accuracy (%)

41
REFERENCES:
[1] Kodituwakku S. R, Selvarajah S. “Comparison of Color Features for Image Retrieval”. Indian Journal
of Computer Science and Engineering, Vol. 1, No. 3, pp: 207-211, 2010.
[2] Felci Rajam I. and Valli S. “SRBIR: semantic region based image retrieval by extracting the dominant
region and semantic learning”. Journal of Computer Science, Vol. 7, No. 3, pp.400–408, 2011a.
[3] Felci Rajam I and Valli S. “Content-Based Image Retrieval Using a Quick SVM-Binary Decision
Tree – QSVMBDT”. Springer Communications in Computer and Information Science 205, pp: 11-22,
2011b.
[4] FelciRajam I. and Valli S. “Region-based image retrieval using the semantic cluster matrix and
adaptive learning”. International Journal of Computational Science and Engineering, Vol. 7, No. 3,
pp.239–252, 2012.
[5] Celia B, Felci Rajam I. “An efficient content based image retrieval framework using machine learning
techniques”. Proceedings of the Second international conference on Data Engineering and
Management (ICDEM ‗10), Springer LNCS, Vol. 6411, pp 162-169, 2010.
[6] Lining Zhang, Lipo Wang andWeisi Lin. “Generalized Biased Discriminant Analysis for Content-
Based Image Retrieval”. IEEE Transactions on Systems, Man, and Cybernetics—Part B: Cybernetics,
Vol. 42, No. 1, pp. 282-290, 2012.
[7]I mtnan-Ul-Haque Qazi, OlivierAlata, Jean-ChristopheBurie, Ahmed Moussa, ChristineFernandez-
Maloigne. “Choice of a pertinent color space for color texture characterization using parametric
spectral analysis”. Pattern Recognition 44, pp. 16–31, 2011.
[8] Hatice Cinar Akakin and Metin N. Gurcan. “Content-Based Microscopic Image Retrieval System for
Multi-Image queries”. IEEE Transactions on Information Technology in Biomedicine, Vol. 16, No. 4,
pp. 758 – 769, 2012.
[9] WangXing-yuan, ChenZhi-feng, YunJiao-jiao. “An effective method for color image retrieval based
on texture”. Computer Standards & Interfaces 34, pp. 31– 35, 2012.
[10] Samuel Rota Bul_o, Massimo Rabbi and Marcello Pelillo. “Content-Based Image Retrieval with
Relevance Feedback using Random Walks”. Pattern Recognition, Vol. 44, No. 9, pp. 2109-2122,
2011.
[11] Neetu Sharma S, Paresh Rawat S and jaikaran Singh “Efficient CBIR Using Color Histogram
Processing. Signal & Image Processing” : An International Journal(SIPIJ), Vol.2, No.1, pp. 94-112,
March 2011.
[12] Ja-Hwung Su, Wei-Jyun Huang, Philip S. Yu, Fellow, and Vincent S. Tseng. “Efficient Relevance
Feedback for Content-Based Image Retrieval by Mining User Navigation Patterns”. IEEE
Transactions on Knowledge and Data Engineering, Vol. 23, No. 3, pp. 360-372, 2011.
[13] Feng S., Lang C. and Xu D. “Localized content-based image retrieval using saliency-based graph
learning framework”. ICSP 2010 Proceedings, pp.1029–1032, 2010.
Authors
A. P.Anand was born on 16th Septemper 1987 in Madurai ,Tamilnadu , India. He
received his B.E degree in Electronics and Communication Engineering at Karpagam
College of Engineering, Coimbatore in 2009 and M.E degree in Communication
Systems at SNS College of Engineering and Technology Coimbatore in 2011. He is
working as Assistant Professor in the Department of Electronics and Communication
Engineering, Velammal College of Engineering and Technology, Madurai. He has 3
years of teaching experience.
T. Ajitha doing her Final year B.E., in Electronics and Communication from Velammal
College of Engineering and Technology, Madurai.

42
M. Priyadharshini doing her Final year B.E., in Electronics and Communication
from Velammal College of Engineering and Technology, Madurai.
M.G. Vaishali doing her Final year B.E., in Electronics and Communication from
Velammal College of Engineering and Technology, Madurai.

Content based image retrieval (cbir) using

More Related Content

What's hot (20)

Viewers also liked (9)

Similar to Content based image retrieval (cbir) using (20)

More from ijcsity (20)

Recently uploaded (20)

Content based image retrieval (cbir) using