Ijaems apr-2016-16 Active Learning Method for Interactive Image Retrieval

International Journal of Advanced Engineering, Management and Science (IJAEMS) [Vol-2, Issue-4, April- 2016]
Infogain Publication (Infogainpublication.com) ISSN : 2454-1311
www.ijaems.com Page | 139
Active Learning Method for Interactive Image
Retrieval
Varsha Hole1
, Bhumi Dave2
, Sumitra Gauda3
, Priyanka Gawas4
1
Professor, Computer Engineering Department, Smt. Indira Gandhi College of Engineering, Mumbai, Maharashtra, India
2,3,4
Student, Computer Engineering Department, Smt. Indira Gandhi College of Engineering, Mumbai, Maharashtra, India
Abstract — With many possible multimedia applications,
content-based image retrieval (CBIR) has recently gained
more interest for image management and web search.
CBIR is a technique that utilizes the visual content of an
image, to search for similar images in large-scale image
databases, according to a user’s concern. In image
retrieval algorithms, retrieval is according to feature
similarities with respect to the query, ignoring the
similarities among images in database. To use the feature
similarities information, this paper presents the k-means
clustering algorithm to image retrieval system. This
clustering algorithm optimizes the relevance results by
firstly clustering the similar images in the database. In
this paper, we are also implementing wavelet transform
which demonstrates significant rough and precise
filtering. We also apply the Euclidean distance metric and
input a query image based on similarity features of which
we can retrieve the output images. The results show that
the proposed approach can greatly improve the efficiency
and performances of image retrieval.
Keywords — CBIR, Clustering, Filtering, Similarity
measures, K-means, Euclidean distance metric.
I. INTRODUCTION
1.1 IMAGE RETRIEVAL
In today’s world the expansion of internet and multimedia
technologies has increased, a huge amount of multimedia
data in the form of audio, video and images has been used
in many fields like medical treatment, satellite data,
digital forensics, surveillance system, etc. This has
created a fragmentary demand of systems that can store
and retrieve multimedia data in an effective way.
In recent years, rapid increase in the size of digital image
database has been seen. Image Retrieval is concerned
with the searching, finding and retrieving of digital image
from a set of images or image database. To search an
image from this large increasing database requires special
concentration. Image Retrieval has been an active
research area since 1970's. A proper structured database
provides effective browsing and searching capability.
Database management and computer vision are two
research communities working on this area. These two
research areas provide two major techniques for searching
an image.
1.2 TEXT BASED IMAGE RETRIEVAL
The most common retrieval systems are Text Based
Image Retrieval (TBIR) systems, where the search is
based on automatic or manual annotation of images. In
Text Based Image Retrieval, text data represents the
image features. Most existing image retrieval systems are
text based. Depending upon the image objects,
characteristics and features some keywords are indexed to
the database to search an image.
A conventional TBIR searches the database for the
similar text nearby the image as given in the query string.
The commonly used TBIR system is Google Images. The
text based systems are fast as the string matching is
computationally less time consuming process. However,
it is sometimes difficult to express the whole visual
content of images in words and TBIR may end up in
producing irrelevant results. In addition annotation of
images is not always correct and consumes a lot of time.
1.3 CONTENT BASED IMAGE RETRIEVAL
For finding an alternative way of searching and
overcoming the limitations imposed by TBIR systems
more spontaneous and user friendly content based image
retrieval systems (CBIR) were developed. A CBIR
system uses visual contents of the images described in the
form of low level features like colour, texture, shape and
spatial locations to represent the images in the databases.
A typical CBIR system is as shown in fig. 1.
Fig. 1 Basic working of CBIR
The system retrieves similar images when an example
image or sketch is presented as input to the system.

Querying in this way eliminates the need of describing the
visual content of images in words and is close to human
perception of visual data.
II. EXISTING SYSTEM
The search for significant information in the large space
of image and video databases has become more
challenging. The main challenge lies in the reduction of
the semantic gap between low-level features extracted
from the image and high-level user semantics. How to
achieve accurate retrieval results are still a challenging
and an unsolved research problem. A typical image
retrieval system includes three major components: 1)
feature extraction (usually in conjunction with feature
selection), 2) high dimensional indexing and 3) system
design. The main contribution of this work is a
comprehensive comparison of colour feature extraction
approaches and texture feature extraction approaches for
CBIR.
2.1 COLOUR FEATURE EXTRACTION MODELS
The extraction of the colour features for each of the four
methods is performed in the HSV (hue, saturation and
value) perceptual colour space, where Euclidean distance
corresponds to the human visual system’s notion of
distance or similarity between colours.
2.1.1 THE CONVENTIONAL COLOUR
HISTOGRAM
The conventional colour histogram (CCH) of an image
indicates the frequency of occurrence of every colour in
the image. From a probabilistic perspective, it refers to
the probability mass function of the image intensities. It
captures the joint probabilities of the intensities of the
colour channels. The CCH can be represented as
hA,B,C(a,b,c) = N. Prob(A=a, B=b, C=c), where A, B and
C are the three colour channels and N is the number of
pixels in the image [3] (key paper #1). Computationally, it
is constructed by counting the number of pixels of each
colour (in the quantized colour space)[1].
2.1.2 THE FUZZY COLOUR HISTOGRAM
The classic histogram is a global statistical feature, which
describes the intensity distribution for a given image. Its
main advantage is that it is fast to manipulate, store and
compare and insensitive to rotation and scale. On the
other hand, it is also quite unreliable as it is sensitive to
even small changes in the scene of the image. In colour
image processing, the histogram consists of three
components, respect to the three components of the colour
space.
A histogram is created by dividing a colour space into a
number of bins and then by counting the number of pixels
of the image that belongs to each bin. It is usually thought
that in order for an image retrieval system to perform
satisfyingly, the number of regions that the colour space
is divided into is quite large and thus the colours
represented by neighbouring regions have relatively small
differences. As a result, the perceptually similar colours
problem appears, images which are similar to each other
but have small differences in scene or contain noise will
produce histograms with dissimilar adjacent bins and vice
versa due to the small distance that the regions are
separated from each other[2].
2.1.3 THE COLOUR CORRELOGRAM
The colour correlogram (CC) expresses how the spatial
correlation of pairs of colours changes with distance. A
CC for an image is defined as a table indexed by colour
pairs, where the dth entry at location (i, j) is computed by
counting number of pixels of colour j at a distance d from
a pixel of colour i in the image, divided by the total
number of pixels in the image [1][3].
Table 1: comparison of colour feature extraction models
Colour
Feature
Pros Cons
Conventional
Colour
Histogram
-Simple
-Fast
computation
-High dimensionality
-No colour similarity
-No spatial information
Fuzzy Colour
Histogram
-Fast
computation
-Encodes colour
similarity
-Robust to
quantization
noise
-Robust to
change in
contrast
-More computation
-Appropriate choice of
membership weights
needed
Colour
Correlogram
-Encodes
spatial
information
-Very slow computation
-Does not encode colour
similarity
2.2 TEXTURE FEARTURE EXTRACTION
MODELS
The notion of texture generally refers to the presence of a
spatial pattern that has some properties of homogeneity.
Directional features are extracted to capture image texture
information. The texture feature extraction methods
presented in this section generate a multi‐scale,
multi‐directional representation of an image.
2.2.1 CONTOURLET TRANSFORM
The Contourlet transform is a combination of a Laplacian
pyramid (LP) and a directional filter bank (DFB). The LP
provides the multi‐scale decomposition, and the DFB
provides the multi‐directional decomposition. The LP is a
decomposition of the original image into a hierarchy of
images, such that each level corresponds to a different
band of image frequencies. This is done by taking the

difference of the original image and the Gaussian low
pass‐filtered version of the image, (at the appropriate
scale σ). The Gaussian low pass kernel is defined as :
H(w1,w2 ) = exp{-2( )2
(w12 + w22)}, where w1 and
w2 are the horizontal and vertical frequencies
respectively. The band pass images from the Laplacian
pyramid are fed into the DFB so that directional
information can be captured. The DFB realizes a division
of the spectrum into 2L wedge‐shaped slices. A detailed
description of the DFB is provided in. The low frequency
components are separated from the directional
components. After decimation, the decomposition is
iterated using the same DFB [1].
2.2.2 THE COMPLEX DIRECTIONAL FILTER BANK
The complex directional filter bank (CDFB) consists of a
Laplacian pyramid and a pair of DFBs, designated as
primal and dual filter banks. The filters of these filter
banks are designed to have special phase functions, so
that the overall filter is the Hilbert transformof the primal
filter bank. A multi‐resolution representation is obtained
by reiterating the decomposition in the low-pass branch
[1][5].
Table 2: comparison of texture feature extraction models
Texture Feature Pros Cons
Contourlet
Transform
-Lower sub-bands
decimated
-Number of
orientations supported
needs to be power of
2
Complex
Directional Filter
Bank
-Competitive
retrieval results
-Computationally
intensive
III. PROPOSED SYSTEM
3.1 SYSTEM DESIGN AND IMPLEMENTATION
STYLE
The System is divided into 5 Modules.
A. Data Base Module
B. Index Module
C. Image Search Module
D. Image Retrieval Module
E. Analysis Module
F. User Interface Module
A. Data Base Module: This Module Maintains data base
as collection of images.
B. Index Module: This Module used to create Index
structure on data base of images.
C. Image Search Module: This Module used to search
similar images based on Query Image.
D. Image Retrieval Module: This Module used to
retrieve similar images based on Query Image.
E. Analysis Module: This Module compares the
performance results on searching the image with index
and without index.
F. User Interface Module: This Module develops user
interfaces for various operations.
Fig. 2 System architecture
3.2 INDEXING
We have used a clustering algorithm, the K-means
clustering algorithm to group the images into clusters
based on the color content. This clustering algorithm has
been frequently used in the pattern recognition literature.
Here we are going to filter most of the images in the
hierarchical clustering and then apply the clustered
images to K-Means, so that we can get better favored
image results.
3.2.1 K-MEANS ALGORITHM
The basic aim is to segment colours in an automated
fashion using the L*a*b* colour space and K-means
clustering. The entire process can be summarized in
following steps:-
Step 1: Read the image Read the image from mother
source which is in .JPEG format, which is a fused image
of part of Bhopal city of Madhya Pradesh, India with
DWT fusion algorithm of Cartosat-1 and LISS-IV of
Indian satellite IRS-P6 and IRS-1D.
Step 2: For colour separation of an image apply the De-
correlation stretching.

Step 3: Convert Image from RGB Colour Space to
L*a*b* Colour Space How many colours do we see in the
image if we ignore variations in brightness? There are
three colours: white, blue, and pink. We can easily
visually distinguish these colours from one another.The
L*a*b* colour space (also known as CIELAB or CIE
L*a*b*) enables us to quantify these visual
differences.The L*a*b* colour space is derived from the
CIE XYZ tri-stimulus values. The L*a*b* space consists
of a luminosity layer 'L*', chromaticity-layer 'a*'
indicating where colour falls along the red-green axis, and
chromaticity-layer 'b*' indicating where the colour falls
along the blue-yellow axis. All of the colour information
is in the 'a*' and 'b*' layers. We can measure the
difference between two colours using the Euclidean
distance metric. Convert the image to L*a*b* colour.
Step 4: Classify the Colours in 'a*b*' Space Using K-
Means Clustering is a way to separate groups of objects.
K-means clustering treats each object as having a location
in space. It finds partitions such that objects within each
cluster are as close to each other as possible, and as far
from objects in other clusters as possible. K-means
clustering requires that you specify the number of clusters
to be partitioned and a distance metric to quantify how
close two objects are to each other. Since the colour
information exists in the 'a*b*' space, your objects are
pixels with 'a*' and 'b*' values. Use K-means to cluster
the objects into three clusters using the Euclidean distance
metric.
Step 5: Label Every Pixel in the Image Using the Results
from K-MEANS For every object in our input, K-means
returns an index corresponding to a cluster.Label every
pixel in the image with its cluster index.
Step 6: Create Images that Segment the Image by Colour.
Using pixel labels, we have to separate objects in image
by colour, which will result in five images.
Step 7: Segment the Nuclei into a Separate Image Then
programmatically determine the index of the cluster
containing the blue objects because Kmeans will not
return the same cluster index value every time. We can do
this using the cluster centre value, which contains the
mean 'a*' and 'b*' value for each cluster. [8]
3.3 SIMILARITY MATCHING
For finding the similarities in the images we use some
distances. These are Euclidean distance, Manhattan
distance, Canberra distance. By using these distances we
find the similarity between images & retrieve most
similar images with query image.
3.3.1 EUCLIDIAN DISTANCE ALGORITHM
In the domain of image retrieval from large databases
using signatures like colour histogram, each ‘n’
dimensional feature vector may be considered as a point
in the ‘n’ dimensional vector space. Thus, a feature vector
is mapped to a point in the n-dimensions. This mapping
helps us to perceive the images (represented by their
feature vectors) as high-dimensional points. The
advantage of this representation is that one can now use
different distance metrics for (i) finding similarity
between two images and (ii) ordering a set of images
based on their distances from a given image. This enables
one to do a nearest neighbour search on a large database
of images and retrieve a result set containing images that
are closest matches to a user-specified query.
It is evident that the images and their ordering depend
both on the feature extraction method as well as on the
distance metric used. In this work, Manhattan distance
and Euclidean distance metrics have been used as a
similarity rule. After the colour, shape or texture
information is extracted from an image, it normally
encoded into a feature vector. Given two feature vectors,
x1 and x2, a distance function computes the difference
between them. It is hoped that this difference will
accurately measure the dissimilarity between the images
from which the features were extracted. The greater the
distance, lesser is the similarity. Distance functions
Euclidean (L2) norm and Manhattan distance or L1 norm
(also known as city block metric) equations are as
follows: The Euclidean distance is given by the following
mathematical expression[4].
, = −
Where, 1, 2 are the coordinates where two
pixels p1 and p2 are located.
The Manhattan distance is given by following
mathematical:
, = −
Distance functions or metrics follows following
properties:
, ≥ 0 , = 0
, = ,
, ≤ , + ,
3.4 FILTERING
Wavelet transform is a convolution operation, which can
equivalent to passing the pixel values of an image through
a low pass filter and a high pass filter. Although
suggested by some researchers and are easier to
implement. Haar wavelets do not have sufficiently sharp
transition and hence are not able to separate different
frequency bands appropriately. Daubechies’ wavelets, on

International Journal of Advanced Engineering, Management and Science (IJAEMS)
Infogain Publication (Infogainpublication.com
www.ijaems.com
the other hand, have better frequency resolution
properties because of their longer filter lengths.
3.4.1 WAVELET TRANSFORM
The transforms are mainly based on small waves, called
wavelets. It shows that both the frequency and temporal
information uses wavelets. Human vision is much more
sensitive to small variations in brightness or colour that is
more sensitive to low frequency signals. To determine the
low frequency area and high frequency area wavelet
transform has been used.Wavelet analysis shown in fig.3
represents a windowing technique with variable
regions. Wavelet analysis allows uses of long time
intervals where we want more precise low frequency
information, and shorter regions where we want high
frequency information [7].
Fig. 3 Wavelet analysis
That other signal analysis technique like breakdown
points,discontinuities in higher derivatives, trends,
similarity also compress as well as de
without appreciable degradation were miss.
Image Decomposition
HL - Horizontal details of the original image.
LH -Vertical details of the original image.
HH - Diagonal details of the original image. (L=Low,
H=High).
HL - Horizontal details of the original image.
LH -Vertical details of the original image.
HH - Diagonal details of the original image. (L=Low,
H=High)
Fig. 4 Image Decomposition
International Journal of Advanced Engineering, Management and Science (IJAEMS)
Infogainpublication.com)
the other hand, have better frequency resolution
properties because of their longer filter lengths.
The transforms are mainly based on small waves, called
. It shows that both the frequency and temporal
information uses wavelets. Human vision is much more
sensitive to small variations in brightness or colour that is
y signals. To determine the
low frequency area and high frequency area wavelet
transform has been used.Wavelet analysis shown in fig.3
represents a windowing technique with variable-sized
regions. Wavelet analysis allows uses of long time
e want more precise low frequency
information, and shorter regions where we want high
like breakdown
higher derivatives, trends, anself-
also compress as well as de-noise a signal
without appreciable degradation were miss.
Horizontal details of the original image.
ge. (L=Low,
Horizontal details of the original image.
Diagonal details of the original image. (L=Low,
ecomposition
In decomposition, firstly image is decomposed
sub, One Level Decomposition. The LL sub
decomposed then called Two Level Decomposition
shown in fig. 4.
IV. EXPERIMENTAL RESULTS
We have followed the image retrieval technique, as
described in the section III. We perfo
mainly on two image sets, such as bus
The following figure explains Indexing creation process.
The clustering is done using the K
clusters can be viewed as shown in fig. 5.
Fig. 5 image search using index
The following figure shows that if there are
instances in the query image as shown in fig. 6
shows that accuracy in the retrieval of the images is
preserved.
Fig. 6 Retrieval results for an
objects
The rough and precise filtering can be
depending upon the user’s interest
Retrieval results shows that the number of images in
precise filtering decreases eventually.
[Vol-2, Issue-4, April- 2016]
ISSN : 2454-1311
Page | 143
on, firstly image is decomposed into four
evel Decomposition. The LL sub-band further
decomposed then called Two Level Decomposition
EXPERIMENTAL RESULTS
We have followed the image retrieval technique, as
described in the section III. We performed experiments
image sets, such as buses and dinosaurs.
The following figure explains Indexing creation process.
The clustering is done using the K-means algorithm,
clusters can be viewed as shown in fig. 5.
image search using index
The following figure shows that if there are different
instances in the query image as shown in fig. 6, result
accuracy in the retrieval of the images is
etrieval results for an image containing more
objects
and precise filtering can be demonstrated
depending upon the user’s interest as shown in fig. 7.
Retrieval results shows that the number of images in
precise filtering decreases eventually.

Fig. 7 Retrieval results showing precise filtering
V. CONCLUSION
Content based image retrieval (CBIR) is a lively
discipline, expanding in both depth and breadth and an
important research topic covering a large number of
research domains like image processing, computer vision,
very large databases and human computer interaction.
One of the main tasks for (CBIR) systems is similarity
comparison, extracting feature signatures of every image
based on its pixel values and defining rules for comparing
images.
Image retrieval algorithms always use the similarity
between the query image and images in image database.
However, they ignore the similarities between images in
image database. In this paper we addressed this problem
by introducing a graph-theoretic approach for image
retrieval post processing step by finding image similarity
clustering to reduce the images retrieving space.
Experiment results show that the efficiency and
effectiveness of k-means algorithm in analysing image
clustering, which also can improve the efficiency of
image retrieving and evidently promote retrieval
precision. This k-means algorithm independent on the
feature extraction algorithm is used as a post-processing
step in retrieval.
REFERENCES
[1] Low-Level Colour and Texture Feature Extraction
for Content-Based Image Retrieval, Final Project
Report, May 09, 2008, EE 381K: Multi-Dimensional
Digital Signal Processing, Michele Saad.
[2] CBIR USING COLOR HISTOGRAM
PROCESSING P.S.SUHASINI, 2Dr. K.SRI RAMA
KRISHNA, 3Dr. I. V. MURALI KRISHNA,
Engineering College, Vijayawada, A.P., India–
520007.
[3] J. Huang, S. R. Kumar, M. Mitra, W. J. Zhu and R.
Zabih, “Image Indexing Using Colour
Correlograms”, Proc. IEEE Conf. on Computer
Vision and Pattern Recognition, pp. 762 – 768, June
1997.
[4] Jasvinder Singh et al Int. Journal of Engineering
Research and Applications, www.ijera.com, ISSN:
2248-9622, Vol. 4, Issue 9( Version 1), September
2014.
[5] S. Oraintara and T. T. Nguyen, “Using Phase and
Magnitude Information of the Complex directional
Filter Bank for Texture Image Retrieval”, Proc.
IEEE Int. Conf. on Image Processing, vol. 4, pp.
61‐64,Oct. 2007.
[6] International Journal of Advanced Research in
Computer and Communication Engineering, Vol. 2,
Issue 5, May 2013 Copyright to IJARCCE
www.ijarcce.com 2181, Content Bases Image
Search And Retrieval Using Indexing By K-Means
Clustering Technique.
[7] International Journal of Innovative Research in
Science, Engineering and Technology, Volume 3,
Special Issue 3, March 2014.2014 International
Conference on Innovations in Engineering and
Technology (ICIET’14).
[8] Application Research of k-means Clustering
Algorithm in Image Retrieval System, Proceedings
of the Second Symposium International Computer
Science and Computational Technology(ISCSCT
’09) Huangshan, P. R. China, 26-28,Dec. 2009, pp.
274-277.

Ijaems apr-2016-16 Active Learning Method for Interactive Image Retrieval

Recommended

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to Ijaems apr-2016-16 Active Learning Method for Interactive Image Retrieval (20)

Recently uploaded (20)

Ijaems apr-2016-16 Active Learning Method for Interactive Image Retrieval