CONTENT BASED IMAGE RETRIEVAL BY CLUSTERING USING (SIFT) Algorithm

tealackingΤεχνίτη Νοημοσύνη και Ρομποτική

8 Νοε 2013 (πριν από 3 χρόνια και 7 μήνες)

62 εμφανίσεις

1


CONTENT BASE
D

IMAGE RETRIEVAL BY
CLUSTERING USING (SIFT) Algorithm


Sambhav jain

dept. Of information technology

svits, sanwer road

Indore, India

sambhav.jain43@gmail.com

Prof. Manoj cho
uhan

dept. Of information technology


svits, sanwer road

Indore, Indi
a

manoj_mits85@yahoo.com


Abstract


CBIR is a technique which uses visual content of an image such as color, textures and shape to search
images from large image databases this is a challenging task. CBIR system becomes much more valuable, this kind of
s
ystems used to assist against various challenges as well as for entertainment purposes; there is a large no. of
techniques which can be used by this system. CBIR had been faces no. of problem’s face like learning of image
similarity, interaction with user,

the need for database, the semantic gap with image features, understanding and find
the key
-
point of image use of k
-
means to make cluster’s to compare similarity. Here we introduce to use of SIFT in
CBIR. SIFT algorithm use in color based and shape based
image retrieval system. This approach relies on the choice
of several parameter which directly impact it’s effectiveness when applied to retrieve images. This paper, we attempt
to evaluate the application of SIFT to refine CBIR and describe some important
role of SIFT in CBIR system,
important conclusion about SIFT algorithm from the process and the result of its implementation.

Keywords
-

CBIR, Clustering, K
-
means, QOM, NNS, SIFT.

I.

I
NTRODUCTION

CBIR is a set of techniques for retrieving semantically relevant

images from the image database.

Nowadays, in
the most of areas it is necessary to work with large amounts of growing visual and multimedia data, at the same
time, the number of image and video files on the web is quite big and is still rising very rapidly
. Searching
through this data is absolutely vital. So, there is a high demand on the tools for image retrieving, which are
based on visual information, rather than simple text
-
based queries. Content
-
based Image Retrieval (CBIR)
consists of retrieving the m
ost visually similar images to a given query image from a database or group of image
files. It is a quite useful thing in a lot of areas such as Photography which may involve image search from the
large digital photo galleries
.

A.

Clustering
:


"clustering" is

essentially a set of such clusters, usually containing all objects in the data set. Additionally, it
may specify the relationship of the clusters to each other, for example a hierarchy of clusters embedded in each
other. Data Clustering is one of the chal
lenging mining techniques exploited in the knowledge discovery
process. Clustering huge amounts of data is a difficult task since the goal is to find a suitable partition in a
unsupervised way (i.e. without any prior knowledge) trying to maximize the simil
arity of objects belonging to
the same and minimizing the similarity among objects in Different clusters Here the “Cluster
-
model” is a Key to
Understanding the Differences b/w Various Algorithm.






Fig1: CBIR BY CLUSTRING

2



B.


SIFT(Scale Invariant Feature Transforms):


SIFT Stand for Scale invariant features transform. The main goal of this is efficiency during image indexing
and retrieval of image, thereby reducing the need for human
interference in the indexing, the computer must be
able to retrieve image from a database without any human assume on specific realm (control). SIFT has Perform
four computational (reckon or calculate) Phases, computations are expensive. Extracting the key
-
point is
minimized by the use of cascading approach, and the output of the SIFT algorithm is a set of key
-
point
descriptors. Once such descriptors have been generated for more than one image.


Four phases are here:





Phase 1: scale
-
space Extrema detection

The first phase of the computation seeks to identify potential interest points. It will searches over all scales and
image locations. The computation is a
ccomplished by using a difference
-
of
-
Gaussian (DOG) function [5]. The
resulting interest points are invariant to scale and rotation, meaning that they are persistent across image scales
and rotation.

The computation is like this:

D(x; y;


(G(x; y; k

)

-

G(x; y;


)) * I(x; y) , (See Fig 3,a)

Phase 2: keypoint localization

For interest point found in phase 1, a detailed model is created to determine location and scale. Keypoint are
selected based on their stability, and a stable keypoint is thus a key
point resistant to image distortion.

Phase 3: orientation assignment

For each of the keypoints identified in phase2, SIFT computes the direction of gradients around. One or more
orientations are assigned to each keypoint based on local image gradient direc
tions.

Phase 4: keypoint descriptor

The local image gradients are measured in the region around each keypoint. These are transformed into a
representation that allows for significant levels of local shape distortion and change in illumination. There are
ni
ne parameters one can assign to adjust what criteria SIFT used on its four
-
step way to identify keypoints

reference number, as in

[1][4]. However, some will only evaluated the performance of a SIFT implementation,
and not looked into exactly how these para
meter affect keypoint generation. The parameters used are the ones
lows set fourth optimal, and his recommendations are based on empirical studies. It is a time to present image
matching, and also getting some result obtained when testing for object recogn
ition in image. “SIFT identified
~4.6 times more key
-
points to other”.




Fig: (3,
a)
and (3,b)

Gaussian function diagram,
keypoint

descriptor







3


II.

P
ROPOSED WORK

SIFT algorithm identifies the features of the query image. These features are compared

with the depending on
these three methods we can calculate the efficiency and reliability of methods using precision and recall basis.
The SIFT algorithm identifies features of an image that are distinct to video and these features can in turn be used
to
identify similar or identical objects in other images. This image goes through four computational phases as
mentioned above. A key point descriptor of given image are produced. A key point is an image feature which is
so distinct that image scaling, noise,

or rotation does not or rather should not distort the key point that is if there is
given a key point in an image, if one scales the image to half the size or double the size, the key point would still
be identifiable. The same goes for image rotation and

noise. If an image is, rotated clockwise, the key point would
still persist. A key point descriptor is a 128
-
dimensional vector that describes a key point. The reason for this high
dimension is that each key point descriptor contains a lot of information
about the point it describes. These key
point features are responsible for retrieval of image. The Search will be carried out as shown in fig (4) below

reference number, as in

[6]. Using SIFT algorithm image as a query produces the good result and solves
t
he
problem of image retrieval.
My
work to search images

in logarithmic and expected time.


Fig. 4

precision vs. numb
er of image retrieve


III.

I
MAGE
M
ATCHING
:


When ones have generated sets of key point descriptors

reference number, as in

[1] for two or more i
mages, one
can begin matching images.

A keypoint descriptor is a 128
-
dimensional vector that describes a keypoint. The
reason for this high dimension is that each keypoint descriptor contains a lot of information about the point it
describes
.






Fig: 5(a) take I/p to find keypoints using SIFT And fig 5(b) to mach two images keypoints using NNS to find which keypoints c
losely match
target image keypoints.

Keypoint
Descriptors
image 1

Keypoint
Descriptor
s
image

2




NNS

Closely
matching

keypoint
Descript
ors

4


IV.

EXPERIMENTAL

RESULT
:

The SIFT algorithm has gone Far reac
hing testing, and here I present results and findings. First of all we store
66 images in database and given numbering them to show result of SIFT algorithm. Make a GUI and put all
operation we use. Pick image one
-
by

one and got keypoints in training time
, Process steps
-

I/p source image
then apply SIFT; find key
-
point descriptors (128 bit) in fig(5.a) . That time searching method which is search
images in database, we use to apply NNS method 1 to n images, after that we found required image and that’s
the

out
-
put. We calculate time with training time; we may also check how much possible cluster use, and also
display some recall image either it is target means single image or two or many image recall, we select an image
from database and search a image on b
ehalf of matching keypoints. It will search image even it is rotate. We
have also calculated time by search time in second, also some time image not find if it is not present in
‘database’. Here we have to show result of matching image with the use of CBIR

using SIFT.

C
ONCLUSION AND FUTURE

WORK

The SIFT does what it is designed to do, and it does it well in my paper is to save time for found key
-
pints and
match them, so here we use NNS to fast possible and target image. In this paper I found some typical im
ages
which have large no. of keypoint then apply difference of Gaussian and refine the keypoints which help to
perform feature matching, some time we got very smooth image and find many features, on that case a small
face could we unrecognized from the tra
ining image. The recognition could performance improve by add new
SIFT feature sizes and offsets.

Future work may be some implementation issues like program language .net, java whatever you preferred for
fast execution to got images, are left as future wor
k.


A
CKNOWLEDGMENT


Here I acknowledge

my work has completed with my
patience
. In this paper
we found result and show
n

with
image an output.

Paper based on SIFT algorithm which is most useful for our work.

R
EFERENCES

[1]

Sambhav jain ’Surv
ey on recent cont
ent base image

retrieval based on Sift algorithm’, ijsws 12
-
334:2013

[2]

Thomas bakken telenor ‘
An evolution of the SIFT

algorithm for CBIR’ R&I N 30/2007.

[3]

Shraddha kumar ‘Radial basis function used in cbir for

Sift features’, ijarcsse : 2012.

[4]

Yu meng ‘Implementing the scale inv
ariant feature

transform method’, computer science.

[5]

yingying ma ‘Dupl
icate images detection based on

SIFT’.

[
7
]

Anil k.jain ‘co
ntent
-
based image retrieval: an

application to tattoo images’, icip 2009.

[8]

Anil balaji

Gonde ‘
SIFT Feature with relevance

feedback for image retrieval’.(issn 0974
-
3375)
.

[9] b.szanto,sami ‘sketch4match


content
-
based imag
e

retrieval system using sketch’, 2011.