Image Retrieval Based On Image Segmemtation Using Integration of Colour and Texture

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

International Journal of Emerging Technologies and Engineering (IJETE)

Volume 3 Issue 6, June 2016, ISSN 2348 8050

IMAGE RETRIEVAL BASED ON IMAGE SEGMEMTATION USING


INTEGRATION OF COLOUR AND TEXTURE
Kavita Sharma
M. Tech. Scholar
Dept. of Electronics & Comm.
Amritsar College of Engg & Technology
Amritsar, Punjab, India

Dr. V. K. Banga
Prof & Principal
Dept. of Electronics & Comm.
Amritsar College of Engg & Technology
Amritsar, Punjab, India

object location and pixels scrawled by the user, which leads to

Abstract
Favorite image extraction from images is technique of finding
images of given objects from image database. The existing
technique uses common template method for extracting image
from database. Common template being created using one
object in both training and target images. In real world case it

the exact extraction of the favorite objects. When the number


of the target images is large then the huge workload will be
produced for the user. Segmentation is an important operation
in image analysis. In real sense image segmentation means to
sub divide an image into its constituent objects.

is uncommon to have a single object in an image unless image


is taken by an expert. This poses a limitation on application of
object extraction technique. We in here will discuss improved
version of image extraction technique by using common

Co-segmentation is the method which we used in this it refers


to the simultaneous segmentation of images. The feasible
method which generates the object prior by introducing the
number of images that contain the favorite objects into the

template technique.

segmentation. It can be extracted by two main ideas. The first


Keywords: Colour, Image Segmentation, image analysis,

method is to generate the common template from the images

texture.

which are introduced and then extract the objects through


matching with the template (common template method). The
another methd is to simultaneously segment the common

1. INTRODUCTION
Image segmentation in general is defined as a process of

objects from the original image and introduced images, which

partitioning an image into homogenous groups such that each

is called co-segmentation. It treats the favorite objects as the

region is homogenous but the union of no two adjacent

common objects and achieves the segmentation through

regions is homogenous. It is a fundamental task in computer

extracting the common objects from a set of images. It can

vision, which can be used in many applications, such as object

handle the shape variation and orientation well.

detection and scene understanding.understanding. In this


paper, we mainly focuson favorite object segmentation which

In this paper, we use common template generation and co-

devotes to segment favorite object from a given image.

segmentation to propose a new favourite object co-

Several existing image segmentation methods [22][10] have

segmentation. We use the common template generation to

been used to extract the favorite object and to obtain the prior

achieve accurate segmentation with low computational cost.

information [10], such as the color, orientation, and the shape

The advantage of the common template method is that the

of the object. A method which is useful to provide the object

generated template can simplify the object segmentation. The

prior information is the user interaction [23], [4], such as the

enhancement has done on image extraction algorithms. The

39
www.ijete.org

International Journal of Emerging Technologies and Engineering (IJETE)


Volume 3 Issue 6, June 2016, ISSN 2348 8050
image extraction algorithms are used to identify objects in

segments to select the successful segments. The selection

image. We have applied orientation and shape of object in

method consists of two steps. The first step is to calculate the

different images to identify images more affectively. The

affinity matrix k= {k }, = 1, ,

enhanced technique has shown more images for same input

the similarity value between any pair of shape features

, = 1, ,

, where k is
and

template. We test the proposed method on many groups of

. Without loss of generality, we set the similarity value in

images, the experimental results demonstrate the effectiveness

the range of [0, 1]. Similar shapes correspond to large

of the proposed method.

similarity value. In the second step, we calculate the eigen


vector and eigenvalue of k, and treat the eigenvector = ((1),

Enhanced Common template generation:

, (

)) corresponding to the largest eigenvalue as the

confidences of the initial segments, i.e., ( ) is the confidence

A. Common Objects Extraction from Training


Images

of initial segmentation p .

We use the saliency value in the calculation of the graph


weight, which can avoid the segmentation of part of the

Along with spectral analysis, RGB and Texture mapping of

common objects. We also add co-segmentation method to

the object is done. For each candidate region found in image

extract the objects.To fit the co-segmentation to our problem,

set rules are run to compare image colors, texture and

many improvements are performed. We will consider that

orientation of the image; For each image

similar shape is shared by the objects, and then we directly use

1. Color comparison: Color comparison is done by extracting

shape feature rather than learning the features to construct the

all colors of the object region. These are then compared to the

graph. Furthermore, for more accurate local region extraction,

colors to objects being extracted from images.

we use multiple local region generation method.

2. Texture matching: Texture of the image is extracted. These


values are then used to compare the texture in test images.

B. Common template generation


There are unsuccessful segments in the segmentation results,
we will select successful segments from the reults of cosegmentation. Then, we align the selected segments to cope
with the variations among the objects. Finally, we generate the

As the successful segments are similar to each other and the


unsuccessful segments are dissimilar to the others, and the
successful segmentation corresponds to large confidence.
Otherwise, it will give small confidence. Hence, unsuccessful
segments can be removed through a threshold. In this paper,

single prototype silhouette.

the threshold is set as the mean value of the eigen vector .

1) Spectral Analysis based Successful Segmentation

2) Favorite Objects Alignment by Directional Shape

Selection:

Representation:

We select successful segmentation by spectral clustering


method. Assume = {p }, = 1, ,

is the set of theinitial

segments (binary mask) by co-segmentation method, where


is the number of the segments.

, = 1, ,

is the shape

Because there are many variations among the shapes of the


common

objects,

such

as

rotation,

translation,

scale

andreflection, we align these shapes before generating the


common template by a new shape descriptor, namely

feature of p . We select the successful segments by the fact

directional shape representation, which can align these shapes

that the successful segments are similar to each other while the

under kinds of variations.

unsuccessful segmentation is dissimilar to the other segments.


Hence, we use the similarity relationships among these

40
www.ijete.org

International Journal of Emerging Technologies and Engineering (IJETE)


Volume 3 Issue 6, June 2016, ISSN 2348 8050

2. Experimental results
Earlier,it has implemented an extraction of the desired objects
from image using common template technique. This requires
the segmenting image and then identifying sections in image.
In existing method, it contain less number of images as shown
in fig.

Fig. 3. The segmentation results using the enhanced favorite object


segmentation method.

The results are shown in Fig where the test images and the
corresponding segmentation results of eight classes are shown.
It is seen from fig that the favorite objects are successfully
segmented from these original images.Here We also
Fig.1: Existing screen contain less images

implement the method in [10] using Matlab. The IOU rate

In fig 2: The enhancement has been done on image extraction

defined as the ratio of the area of the intersection regions to

algorithms. The image extraction algorithms are used to

the area of the union of the ground truth and the obtained

identify objects in image. we have applied orientation and

region is used for the objective evaluation. For accurate

shape of object in different images to identify images more

segmentation we need a large value of IOU. Since there are

affectively. In enhanced method, it contain more images.

more than two images for each class, we use the mean IOU for
verification of each class. The IOU values and the proposed
method are shown in Table I. These IOU are obtained under
the same training images and testing images for comparison.
In
Table I we can see that the enhanced method which can
improve the object segmentation in these groups.

Table1: Comparison with the existing method and the


enhanced method

Fig.2: Enhanced screen contain more images

We collect images from the databases such as icoseg

method

applelogo

cow

heart

moon

swans

existing

0.4466

0.7892

0.6312

0.853

0.6595

enhanced

0.4466

0.80001

0.6812

0.893

0.6895

databases, MRSC database, Caltech database and ETHZ shape


database verification. The groups collected form web are also
considered (heart and moon).

3. CONCUSION AND FUTURE WORK


In this paper we segment favorite object from natural images
very efficiently. The method consists of steps of co-

41
www.ijete.org

International Journal of Emerging Technologies and Engineering (IJETE)


Volume 3 Issue 6, June 2016, ISSN 2348 8050
of

Image

Segmentation

[9]H.P.

extraction. We verify the proposed method by images

Techniques International Journal of Science and Modern

collected from the image datasets and web. The experimental

Engineering (IJISME) ISSN: 2319-6386, Volume-1, Issue-8,

results show that our method can segment the favorite object

July 2013.

from the natural images very accurately.In future this

[10]Hongliang Li, K.N. Ngan, and Qiang Liu.Faceseg:

technique

Automatic facesegmentation for real-time video. IEEE

can

be

used

for

overlapping

objects

by

Narkhede

Review

saliency,co-segmentation, template generation and object

approximating shape of a object that is overlapped by another

Transactions on Multimedia, 11(1):7788, jan. 2009.

object and will focus on extracting partial objects by making

[11]JamilA.M.Saif, Ali Abdo Mohammed, Ali-Kubati Image

minor changes in the code.

Segmentation Using Edge Detection and Thresholding13th


international arab conference on information technology
ACIT,Dec 2012.

REFERENCES
and

[12]Jagdish Prasad goswami, NamartaSaxenaFace Detection

Development of Systems for Image Segmentation and Content

Using color Based Segmentation and Edge Detection

Based Image Retrieval2012 IEEE.

International journal of computer applications,Vol-3 6th june-

[2]A.H.Kam, T.T.Ng, N.G.Kingsbury and W.J.Fitzgerald

2013.

Content Based Image Retrieval through Object Extraction

[13]JagadeeshD.Pujari,

and Querying2000 IEEE.

Filtering on the Performance of Segmentation Based CBIR

[3]B.G. Prasad, K.K. Biswas, and S.K. Gupta Region-based

System International Conference on Signal and Image

image retrieval using integrated color, shape, and location

Processing, 2010 IEEE.

indexComputer Vision and Image Understanding, 2004.

[14]JifengNing,Lei Zhang, David Zhang Interactive image

[4]ChinkiChandhokColor Image Segmentation Using K-

segmentation by maimal similarity based region merging

Means ClusteringInternational Journal of VLSI & Signal

pattern recognition, 2nd , March-2009.

Processing Applications, Vol 2,3rd june2012.

[15]K. K. Singh, A. SinghA Study of Image Segmentation

[5]Chaobing Huang Color Image Segmentation by Seeded

Algorithms for Different Types of Images International

Region Growing and Region MergingSeventh International

Journal of Computer Science Issues, Vol.7, Issue5, 2010.

Conference on Fuzzy System and Knowledge Discovery, 10-

[16]Nikhil R Pal and Sankar K Pal A review on Image

12,Aug-2010.

Segmenatation Techniques, Indian Statistaical Institute,

[6]FanmanMeng, Bing Luo, Chao Huang, Liangzhi Tang,

Pattern Recognition,Vol.26,No.9,pp.1277,93.

[1]Amol

Bing

Bhagat,

Mohammad

Favourite

Zen

Object

AtiqueDesign

Extraction

Using

Web

AratiS.NayakEffect

of

Region

[17]P.S. Hiremath, Jagadeesh PuIari Enhancement Of Image

Images2014 IEEE.

Retrieval By Image Attention Based On Edge Features In a

[7]FanmanMeng, Hongliang Li, Guanghui Liu, and King

Multiresolution

NgiNganObject co-segmentation based on shortest path

Conference on Signal Processing and Communications

algorithm

(ICSPC 2007), 24-27 November 2007.

and

saliency

modelIEEE

Transactions

on

Grid

Framework

IEEE

International

Multimedia, Oct. 2012.

[18] Rastgarpour M., and Shanbehzadeh J.: Application of AI

[8]Gunhee Kim, Eric P. Xing, Li Fei-Fei, and Takeo

Techniques in Medical Image Segmentation and Novel

KanadeDistributed

cosegmentation

submodular

Categorization of Available Methods and Tools, Proceedings

optimization

anisotropic

diffusionInternational

of the International MultiConference of Engineers and

on

Conference on Computer Vision, 2011.

via

Computer Scientists 2011 Vol I, IMECS 2011, March 16-18,


2011.

42
www.ijete.org

International Journal of Emerging Technologies and Engineering (IJETE)


Volume 3 Issue 6, June 2016, ISSN 2348 8050
[19]Rajeshwar

Dass,

Priyanka,

Swapna

Devi

Image

[26]Vittorio

Ferrari,

Frederic

Jurie,

and

Segmentation Techniques International Journal of Electronics

CordeliaSchmidAccurate object detection with deformable

& Communication Technology, Vol. 3, Issue 1, Jan. - March

shape models learnt from imagesIEEEConference on

2012.

Computer Vision and Pattern Recognition, 2007.

[20]Schiewe,J., Segmentation of high-resolution remotely

[27]V. K. Dehariya, S. K. Shrivastava, R. C. JainClustering

sensed

In:

of Image Data Set Using K-Means and Fuzzy K-Means

TheInternational Archives of the Photogrammetry, Remote

Algorithms, International conference on CICN, pp. 386- 391,

Sensingand Spatial Information Sciences, Ottawa, Canada,

2010.

Vol. XXXIV, Part 4, pp. 358-363,2002

[28]Wahba Marian, An Automated Modified Region Growing

[21]ShaiBagon, OriBrostovski, Meirav Galun, and Michal

Technique

Irani Detecting and sketching the commonIEEE Conference

Ultrasound Images, Masters Thesis, Department of Electrical

on Computer Vision andPattern Recognition, 2010.

and Computer Engineering, University of Waterloo, Waterloo,

[22]T. Nguyen, JianfeiCai, Juyong Zhang, and Jianmin Zheng.

Ontario, Canada, 2008.

Robustinteractive image segmentation using convex active

[29]W.X. Kang, Q. Q. Yang, R. R. Liang,The Comparative

contours.

Research

data-concepts,

IEEE

applications

Transactions

on

and

problems.

Image

Processing,

21(8):37343743, 2012.
[23]Tohid

Sedghi,

for

on

Prostate

Image

Segmentation

Segmentation

in

Trans-Rectal

Algorithms,

IEEE

Conference on ETCS, pp. 703-707, 2009.

Majid

Fakheri,

and

Mahrokh

G.

[30]Y.

Zhang,

H.

Qu,

Y.

Wang,Adaptive

Image

ShayestehRegion and Content Based Image Retrieval using

Segmentation Based on Fast Thresholding and Image

Advanced Image Processing Techniques 2010 IEEE.

Merging, Artificial reality and Telexistence-Workshops, pp.

[24]Tanveer

308-311, 1994.

Syeda-Mahmood,

Dragutin

PetkovicOn

describing color and shape information in images 2000

[31]Ying Nian Wu, Zhangzhang Si, Haifeng Gong, and Song-

Elsevier Science

Chun Zhu Learning active basis model for object detection

[25]T.

Lindeberg

and

M.-X.

Li

"Segmentation

and

and recognition International Journal of Computer Vision,

classification of edges using minimum description length

2010.

approximation and complementary junction cues", Computer

[32]Zhang, Y. J:An Overview of Image and Video

Vision and Image Understanding, vol. 67, no. 1,pp.88--

Segmentation in the last 40 years, Proceedings of the 6th

98,1997.

International Symposium on Signal Processing and Its


Applications,pp.144-151,2001.

43
www.ijete.org

You might also like