More Related Content
Similar to A New Approach for CBIR – A Review (20)
More from IRJET Journal (20)
A New Approach for CBIR – A Review
- 1. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 11 | Nov -2017 www.irjet.net p-ISSN: 2395-0072
© 2017, IRJET | Impact Factor value: 6.171 | ISO 9001:2008 Certified Journal | Page 1969
A New Approach for CBIR – A Review
Harkamal Kaur1, Er. Manit Kapoor2, Dr. Naveen Dhillon3
1,2,3 Department Of ECE, IKPTU,Ramgaria College of Engineering and Technology, Phagwara
------------------------------------------------------------------------***---------------------------------------------------------------------
Abstract— The content based image retrieval (CBIR)
methods are used to discover the similar images in
accordance with the input image from the database. The
image retrieval applications plays a vital role in the case
of big databases, where thousands of millions or more
images are stored. In the case of media sharing platforms
such as Instagram, Whatsapp, Facebook, Picasa, etc, a
very large number of data is uploaded on these portals on
the daily basis, which makes it impossible to discover the
relevant image data manually. Hence there is a strong
requirement of versatile information based image
retrieval engines from such databases, which can discover
the relevant images out of the given database. In this
paper, an innovative model for the image retrieval on the
basis of color and texture features has been proposed,
which is expected to resolve the issue related to the
accuracy of image retrieval engines. The performance of
the model would be analyzed by using the accuracy
metrics such as recall, precision, F1-measure and overall
accuracy.
Keywords: CBIR, image processing, visual features,
texture features
INTRODUCTION
Cloud computing is the delivery of computing services
over the Internet. Cloud services allow individuals and
businesses to use software and hardware that are
managed by third parties at remote locations. Examples
of cloud services include online file storage, social
networking sites, webmail, and online business
applications. The cloud computing model allows access
to information and computer resources from anywhere
that a network connection is available. Cloud computing
provides a shared pool of resources, including data
storage space, networks, computer processing power,
and specialized corporate and user applications. The
characteristics of cloud computing include on-demand
self-service, broad network access, resource pooling,
rapid elasticity and measured service. On-demand self-
service means those customers (usually organizations)
can request and manage their own computing resources.
Broad network access allows services to be offered over
the Internet or private networks. In remote data centers,
customers have choice to draw the resources from a pool
of computing resources. The number of services can be
either small or large; and use of a service is measured
and customers are billed accordingly.
The service models of cloud computing can be classified
as: Software as a Service i.e. SaaS, Platform as a Service
i.e. PaaS and Infrastructure as a Service i.e. IaaS. In
Software as a Service model, a pre-made application,
along with any required software, operating system,
hardware, and network are provided. In PaaS, an
operating system, hardware, and network are provided,
and the customer installs or develops its own software
and applications. The IaaS model provides just the
hardware and network; the customer installs or
develops its own operating systems, software and
applications. While there are benefits, there are privacy
and security concerns too. Data is travelling over the
Internet and is stored in remote locations. In addition,
cloud providers often serve multiple customers
simultaneously. All of this may raise the scale of
exposure to possible breaches, both accidental and
deliberate. Concerns have been raised by many that
cloud computing may lead to “function creep” uses of
data by cloud providers that were not anticipated when
the information was originally collected and for which
consent has typically not been obtained. Given how
inexpensive it is to keep data, there is little incentive to
remove the information from the cloud and more
reasons to find other things to do with it.
Figure 1: Framework of encrypted cloud data to retrieve
the files based on similar search
The need to segregate data when dealing with providers
that serve multiple customers, potential secondary uses
of the data—these are areas that organizations should
- 2. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 11 | Nov -2017 www.irjet.net p-ISSN: 2395-0072
© 2017, IRJET | Impact Factor value: 6.171 | ISO 9001:2008 Certified Journal | Page 1970
keep in mind when considering a cloud provider and
when negotiating contracts or reviewing terms of service
with a cloud provider
Given that the organization transferring this information
to the provider is ultimately accountable for its
protection, it needs to ensure that the personal
information is appropriate handled. The two factors are
analyzed as a solution in the existing system which are
related to search privacy requirement i.e. keyword
privacy and file confidentiality
1. File confidentiality: Since then, file content have to be
processed, thus the strength of the file confidentiality
heavily depends upon security strength of symmetric
encryption.
2. Keyword privacy: During the symmetric encryption
scheme, the query trapdoor was generated so the
privacy of query keyword depends on the security
strength of the symmetric encryption scheme.
CBIR TECHNIQUES
There exist several techniques to retrieve the images but
there exist problem of retrieving the images on the basis
of pixels.
Semantic Retrieval: When user makes requests like “find
images of Barack Obama” then semantic search is
started. But this task is very difficult to perform by
computers. Therefore lower level features like color,
shape and texture are used. The results of image
retrieval also require human feedback to identify the
higher level concepts.
Relevance Feedback: In order to make the use of CBIR
successful there is need to understand the ability of user
intent. CBIR make use of relevance.
Feedback: where users mark the resulted images as
relevant or not relevant or neutral and then replace the
search image with the relevant new information.
Other query methods: These may include methods like
image retrieval by image region, by visual sketch, by
direct specification of image features, by touch, voice etc.
Image Distances Measures: Two images can be
compared on basis of their distance measures. Various
dimensions of images are used such as color, texture,
shape and others. The distance of value 0 indicates exact
match with image query. Thus the results are then
stored on basis of their distances to the queried image.
Figure2 Content Based Image Retrieval
Color: Method of image retrieval in this technique is
based on the measure of color similarity by computing
color histogram for each image that signifies the
proportion of pixels of an image. This is the most widely
used technique because it can be performed without
regard to image size. Color proportions are further
classified on the basis of region and spatial relationship
among several color regions.
Texture: This method spatially defines the image and
also looks for visual patterns. Depending on the number
of textures detected in an image, they are represented
as” texels” and then placed into number of sets. This
defines the location of texture. Texture is identified by
modeling it in a two dimensional gray level variation.
Methods to classify textures are co-occurrence matrix,
laws texture energy and wavelet transform.
Shape: Shape doesn’t consider whole image but to shape
of a particular region to be sought out. Two processes
are applied first segmentation or edge detection to
image. Shape filters and shape descriptors are also used.
Some shape descriptors include Fourier transform and
moment invariant
USER FEEDBACK TECHNIQUES FOR CBIR
Relevance feedback based interactive retrieval approach
takes into account the two distinct characteristics of
CBIR, first is the gap which exist between the high level
concepts and low level features of the image and second
is the subjectivity of human perception of visual content.
Thus during the retrieval process both the
characteristics are captured by dynamically updated
weights that are based on the user’s relevance feedback.
In other words we can say that it is used to increase the
accuracy of the image being searched. One of the
- 3. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 11 | Nov -2017 www.irjet.net p-ISSN: 2395-0072
© 2017, IRJET | Impact Factor value: 6.171 | ISO 9001:2008 Certified Journal | Page 1971
methods to retrieve images which are used to compute
the local feature relevance is PFRL method. The top N
results are shown to the end user when some input
query image is given. Now feedback is required from
user side to select the images which are relevant to the
query image. Here the images are classified in two
sections one is containing relevant images and the other
is containing dissimilar images, then the average is
calculated of the two sections. In case all the images are
discarded, the set of new images is selected from
database. The process is continued until the user gets the
desired images.
LITERATURE SURVEY
Song et al. [4] proposed the cryptographic methods for
the problem of searching over encrypted data and
provided the security proofs for the resulting crypto
systems. Techniques have several crucial advantages.
They are probably more secure: they provide provable
secrecy for encryption, means that the un-trusted server
cannot draw anything about the plaintext when only
cipher-text is given. Also the un-trusted server cannot
learn anything more about the plaintext but only the
search result, meaning that they provide query isolation
for searches. They provide controlled searching means
without the user's authorization, the un-trusted server
cannot search for an arbitrary word. They also provide
users the facility of hidden queries, so that they may ask
the un-trusted server to search for a secret word without
revealing that word to the server. Curtmola et al. [2]
presented a per-keyword index construction, where each
entry of the table represent the whole hash table index
which contains the trapdoor for a keyword and an
encrypted set of file identifiers. According to this
searchable symmetric encryption scheme a party is
allowed to outsource the storage of its data to another
party in a private manner and maintaining the ability to
search over it selectively. Wang et al. [3] proposed that
for the first time they formalize and solve the problem of
effective fuzzy keyword search over encrypted cloud
data as well as maintain the keyword privacy. Fuzzy
keyword search is greatly used to enhance system
usability by returning only the matching files when
users' searching inputs exactly match the predefined
keywords or the closest possible matching files based on
keyword similarity semantics, when exact match fails.
Wang et al. [5] proposed a solution for ranked single-
keyword search regarding the certain relevance score.
For the first time this paper define and solve the
problem of secure ranked keyword search over
encrypted cloud data. Ren et al. [6] suggested the similar
secure per-file index, where for each file an index
including trapdoors of all unique words is constructed.
Here, the author proposed several critical security
challenges and suggested for future investigation of
security solution for a trustworthy public cloud
environment. Cao et al. and Yang et al. [1, 8] proposed a
scheme for multi-keyword ranked search, where inner
product similarity is used for result ranking. This paper,
for the first time, defines and solves the challenging
problem of privacy preserving multi-keyword ranked
search over encrypted cloud data. Xia et al. [7] described
that the results could return not only the exactly
matched files, but also the files including the terms
which are semantically related to the query keyword.
Thus in the proposed scheme, a corresponding file
metadata is constructed for each file. Now both the
encrypted metadata set and file collection are uploaded
to the cloud server. With the help of metadata set, the
cloud server builds the inverted index and constructs
semantic relationship library (SRL) for the keywords set.
After receiving a query request, the cloud server first
finds out the keywords which are semantically related to
the query keyword according to SRL.
PROBLEM FORMULATION
Cloud data retrieval or search is the process of the
searching the similar search data against the user query
submitted in the form of image or text. Existing search
data retrieval algorithm in the search work [7] supports
one keyword queries only. The existing technique in the
search work [7] is based on the search method over the
encrypted cloud data. The major point is that the data
over cloud platforms is generally stored in the encrypted
form to ensure the data security, which increases the
response time, which is big problem in searching
process. The researcher in the existing technique has
proposed the semantic user search which avoids the
repeated search results, which are unwanted by the user.
The resultant array sizes in the existing project are
usually kept broad and carry a lot of search data
according the relevance/similarity with search query
image or text, out of which the most of the results are
irrelevant to the user’s search. These results appear
again and again in the similar searches and make the
selection of the relevant search results difficult for the
user. According to our literature and feasibility study,
the solution to eliminate these irrelevant searches can be
developed using multi-layered data matching and user’s
interaction to facilitate the semantic features, which is
capable of remembering the user’s choice to eliminate
the selected irrelevant results from the search data and
will facilitate the robust search than the existing project.
PROPOSED MODEL
The encrypted cloud data storage will be simulated using
MATLAB simulator. The image or text data have to store
as the cloud data in the proposed search work. In the
next phase, the data search algorithm FSRM (Fuzzy
- 4. International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 04 Issue: 11 | Nov -2017 www.irjet.net p-ISSN: 2395-0072
© 2017, IRJET | Impact Factor value: 6.171 | ISO 9001:2008 Certified Journal | Page 1972
Semantic Relevance Matrix) using single keyword search
would be implemented as search technique. The FSRM
algorithm will used compact semantic feature to
remember the unwanted results and such unwanted
results will not appear again in the next search. The
response time of encrypted data should be less by using
this method. The cloud data retrieval is the process of
searching the data over clouds by submitting the custom
search query In the form of text or image. We will
demonstrate the proposed algorithm with the images
stored over the cloud platform using the MATLAB
environment. Our proposed system will use multi-layer
image retrieval algorithm with the semantic feature
enabled for multi-keyword search queries. The first layer
will search the images on the basis of various image
unique properties and/or low-level image features on
the layer-to-layer architecture. Fuzzy Semantic
Relevance Matrix (FSRM) will be used to provide the
semantic features to the cloud platform to avoid the
repeated unused/unwanted results appearing in the
user search results.
CONCLUSION
In this paper on querying an image, a reduced set of
candidate images. The color histogram for an image is
constructed by quantizing the colors within the image
and counting the number of pixels of each color. The
feature vector of an image can be derived from the
histograms of its color components and finally can set
the number of bins in the color histogram to obtain the
feature vector of desired size. Fuzzy relevance semantic
matrix is applied to the relevance feedback of image
retrieval, According to the user’s feedback, to adjust the
weight of FSRM, to catch the user’s intension. After the
limited training, the weight of each of the image class
FSRM modified according to the algorithm in this paper,
thus, there is a good result in the more feedback times.
The algorithm is similar to the experience of mechanism
of human brain and has an initial learning mechanism.
Experiment results clearly show the effectiveness of the
algorithm.
[2] Curtmola, R., Garay, J., Kamara, S., & Ostrovsky, R.
(2006, October), “Searchable symmetric encryption:
improved definitions and efficient constructions”, In
Proceedings of the 13th ACM conference on Computer
and communications security, pp. 79-88.
[3] Li, J., Wang, Q., Wang, C., Cao, N., Ren, K., & Lou, W.
(2010, March), “Fuzzy keyword search over encrypted
data in cloud computing”, In INFOCOM, 2010
Proceedings IEEE, pp.1-5.
[4] Song, D. X., Wagner, D., & Perrig, A. (2000), “Practical
techniques for searches on encrypted data”, In Security
and Privacy, 2000. S&P2000 Proceeding pp. 44-55.
[5] Wang, C., Cao, N., Li, J., Ren, K., & Lou, W. (2010, June),
“Secure ranked keyword search over encrypted cloud
data”, In Distributed Computing Systems (ICDCS), 2010
IEEE 30th International Conference , pp. 253-262.
[6] Wang, C., Cao, N., Ren, K., & Lou, W. (2012). Enabling
secure and efficient ranked keyword search over
outsourced cloud data. Parallel and Distributed Systems,
IEEE Transactions , pp. 1467-1479.
[7] Xia, Z., Zhu, Y., Sun, X., & Chen, L. (2014). Secure
semantic expansion based search over encrypted cloud
data supporting similarity ranking. Journal of Cloud
Computing, pp. 1-11.
[8] Yang, C., Zhang, W., Xu, J., Xu, J., & Yu, N. (2012,
November). A fast privacy-preserving multi-keyword
search scheme on cloud data. In Proceedings of the 2012
International Conference on Cloud and Service
Computing , pp. 104-110
REFERENCES
[1] Cao, N., Wang, C., Li, M., Ren, K., & Lou, W. (2014),
“Privacy-preserving multi-keyword ranked search over
encrypted cloud data”, Parallel and Distributed Systems,
IEEE Transactions , pp. 222-233.