A Survey on Image retrieval techniques with feature extraction

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395 -0056
Volume: 03 Issue: 01 | Jan-2016 www.irjet.net p-ISSN: 2395-0072
© 2016, IRJET ISO 9001:2008 Certified Journal Page 323
A Survey on Image retrieval techniques with feature extraction
Mr. S.S. Mukati 1, Mr. Nitesh Rastogi 2
1 PG Scholar, Computer Science and Engineering Department, Indore Institute of Science and Technology Indore ,
M.P., India
2 Assistant Professor, Computer Science and Engineering Department, Indore Institute of Science and Technology
Indore , M.P., India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - In the past researches, Image retrieval was
the part of searching trends in research and
development. The reason behind those trends is
additional need of multimedia and text data on
internet. As the content based image retrieval
techniques outperforms because of its usability over
syntactical features and their content based
comparisons. But that is more promising to work in
direction of finding more accurate images from their
description or their tags. Therefore this paper discusses
different techniques and methods that are recently
developed for improving the performance of image
search. In addition of that a new model is suggested so
far to optimize the performance of the existing
techniques.
Key Words: CBIR, survey, tag based, feature
computation, system modelling
1. INTRODUCTION
Content Based Image Retrieval (CBIR) [1] is a scheme that
searches images from a large database by means of visual
contents, as per the interest of user. Since 1990s, it has
been rapid growing research area. Moreover, in the past
years, the researchers have made notable results. In late
1970s, the field got its foundation in a conference on
Database Technical for Pictorial Applications which was
held in Florence [3]. This made the researchers to be
attracted towards the field. At early stage, the technique
was based on textual annotations of images i.e. images
were first annotated with text and then searched using a
text-based scheme from typical database systems [2]. This
scheme was little simple and could sometimes fail to
deliver precise results. Moreover, it is not easy to
automatically generate annotations for each image,
therefore manual annotation was followed which is a
clumsy and complicated task plus much expensive if we
have bulk databases.
Further in early 1990s, with rapid growth in internet and
digital image sensors, usage and production of images
increased which further created a need of CBIR systems.
Since then, research on content-based image retrieval has
developed rapidly [4].
This article lists out some essential works and
contributions place in the direction of CBIR and presents a
brief survey about different techniques. In addition of that
the work is extended and obtained a new image retrieval
model. The detailed discussion of the image retrieval
model is given in further sections.
2. BACKGROUND
This section provides the basic understanding of the
content based image retrieval and their functioning.
2.1. Image Retrieval
Image retrieval systems with their basic components are
simulated using figure 1. That can be identified as
searching, browsing, and retrieving images from gigantic
databases consisting of digital images. Although
Traditional and ordinary techniques of retrieving images
employs adding metadata i.e. captioning keywords so as
to perform annotation of words. However image search
can be portrayed by dedicated technique of search which
is habitually used to find images. For that, user provides
the query image and the system returns the image similar
to that of query image [5].
Fig -1: General Image Retrieval System

The image Retrieval has been adopted by most of the
major search engines, including Google, Yahoo!, Bing, etc.
Surrounding texts is considered by a large number of
search engines and image names to index them, since
there are only two main places where text can be placed.
First is title and the other is caption or tags which are
proposed and implemented using web 2.0 concepts.
Usually users make query in the text format for search
contents over any search engine.
2.2. Image Retrieval Techniques
With the expanding use of internet and availability of
different image capturing devices such as digital camera,
bulk amount of images are being created every day in
different areas including remote sensing, fashion, crime
prevention, publishing, medicine, architecture, etc.
Therefore, development of efficient and effective
methodologies to manage large image databases for
retrieval is urgently needed. Three methods of image
retrieval are used i.e. text-based method, content-based
method and hybrid method. This section explains in
details each method. Image retrieval system can be
classified in two key streams as given in figure 2:
• Text based Image retrieval system
• Content Based Image retrieval system
Fig -2: Image Retrieval Approaches
2.3. Text-based image retrieval
Text Based Image Retrieval (TBIR) is presently used in
nearly all general-purpose web image retrieval systems
nowadays. This approach uses the text allied with an
image to find out what the image contains. This text can be
text surrounding the image, the image's filename, a
hyperlink leading to the image, an annotation to the image,
or any other piece of text that can be associated with the
image. Examples of systems using this approach are
Google, Yahoo Image Search engines. These search engines
having indexed over one billion images. Although after
being fast and robust, these search engines sometimes fail
to retrieve relevant images.
2.4. Content - based image retrieval
Content Based Image Retrieval is a set of techniques for
retrieving semantically-relevant Images from an image
database based on automatically-derived image features
[6]. CBIR seeks to avoid the use of textual query inputs.
Rather, it retrieves images based on their visual similarity
to a user-supplied query image or user-specified image
features.
The core objective of CBIR is efficiency while retrieval and
indexing of image, thereby reducing the need for human
intervention in the indexing process [7]. The computer
must be capable of retrieving images from a database
without any human assumption on specific domain (such
as texture vs. non texture). One of the key tasks for CBIR
systems is similarity comparison, extracting feature of
every image based on its pixel values and defining rules
for comparing images. These features become the image
representation for the measurement of similarity with
other images in the database. Images are compared by
calculating the difference of its feature components to
other image descriptors. These image descriptors are
globally obtained on the basis of color histograms for color
features; global texture information on coarseness,
contrast, and direction; and shape features about the
curvature, moment’s invariants, circularity, and
eccentricity.
Majority of existing CBIR systems takes each image as a
complete; although, a single image may comprise multiple
regions/objects with totally different semantic meaning.
An ordinary user will often be interested in only one
particular region of the query image not the whole. So,
instead of viewing an image as whole, viewing it as a set of
regions is more reasonable. Most Image Retrieval systems
include colour, texture, shape and spatial layout. Such
feature lose their effectiveness for CIBR if they get
retrieved from a whole image, because they suffer from
distinct backgrounds, overlaps, occlusion and cluttering in
different images and they do not possess satisfactory
capability to capture mandatory properties of objects, as a
consequence the primary focus of most popular
approaches is being shifted from the global content
description of images into the local content description by
regions or even the objects in images. RBIR is a showing to
be a potential extension of the classical CBIR: rather than
using global features over the entire content, RBIR
systems partition an image into a number of homogenous
regions and extract local features for each region then
features of regions are used to represent and index images
in RBIR. For RBIR, The user puts in a query object by
selecting a region of a query image and then the
consequent similarity measure is computed between
features of region in the query and a set of features of
segmented regions in features database and the system
returns a ranked list of images that contain the same
object. The content-based approach can be summarized as
follows:
1. Computer vision and image processing techniques
are utilized to extract content features from the image.

2. Images are represented as collections of their
distinguished features. For a given feature, an appropriate
representation of the feature and a notion of similarity are
determined.
3. Image retrieval is performed based on computing
similarity or Dissimilarity in the feature space, and results
are ranked based on the similarity measure.
3. RELATED WORK
This section provides the recently made contributions and
efforts in the direction of content based image retrieval
technique.
In this day and age, Content based image retrieval (CBIR)
has become a stronghold of image retrieval systems. In
order to get more yields, a feature of relevance feedback
was added to CBIR so that more precise results can be
obtained by taking user’s feedback into considering.
Though, iterative feedbacks are being offered by existing
systems to obtain refined results, particularly in systems
with large scale image database. Unfortunately this is
infeasible in real applications. In 2011, Ja-Hwung
proposed a novel approach, Navigation-Pattern-based
Relevance Feedback (NPRF), to cope up with large scale
image database thereby improving efficiency and
effectiveness of CBIR. With use of navigation patterns
extracted from user query log, iterations are reduced. As
consequences, efficiency is increased. This proposed
algorithm utilizes the extracted navigation patterns and
three varieties of query refinement strategies named
Query Point Movement (QPM), Query Reweighting (QR),
and Query Expansion (QEX), to congregate the search
space towards the specification of user. By using NPRF
method, high quality of image retrieval on RF can be
accomplished in a slight number of feedbacks [8].
In addition, texture features such as the entropy based on
the gray level co-occurrence matrix and the edge
histograms of an image are also taken into consideration.
Moreover, the IGA is employed to bridge the gap between
retrieved results and user’s expectation and help him to
get the image most satisfying user’s necessity.
Experimental results clearly demonstrate feasibility of the
method [9].
CBIR systems are having their hands in so many different
fields and domains. A CBIR based image retrieval system
was developed for medical applications by Ramesh in
2012. In this application the CBIR is used for searching
medical image collections in large scale. The system
should be capable of retrieving images from the same
disease class as the patient is suffering. The study
presented by Ramesh examines the extraction of
histogram from a medical image. That image is resample
and classified by using Sequential minimal optimization
technique for various percentage of dataset produced
subsequent to the extraction [10].
A number of variants of relevance feedback are available
these days. In order to find out most optimum
methodology, an experiment was presented by
Ghanshyam in 2012 in which various new applications of
genetic algorithm to information retrieval, most of them
related to relevance feedback. Efficient storage and
retrieval of image data is always mandatory in order to
perform assigned tasks and making decisions. The
manuscript presented a new framework for image
retrieval with two types of relevance feedback named
implicit feedback and explicit feedback. It utilizes
Interactive Genetic Algorithm to discover a combination of
descriptors that better characterizes the need of user in
terms of image resemblance [11].
P. Jayaprabha has presented a study in similar domain, in
2013. They presented a high level semantic retrieval
process, in which the search engine is utilized for
retrieving a large number of images using a given text
based query. In a low level image retrieval process, similar
image search function is provided for user to fill in the
input query for image similarity characterizations. The
internet revolution and digital technologies have obligated
a need to have a system to organize abundantly available
digital image for effortless categorization and retrieval.
The techniques involves broad areas, i.e. image
segmentation, image feature extraction, representation,
mapping of features to semantics, storage and indexing,
image similarity distance measurement and retrieval
making CBIR system development a exigent task [12].
Local Binary Patterns are extensively used for texture
classification, have projected a method for facial
expression recognition using Local Binary Patterns (LBP)
as features and Artificial Neural Network [13]. Six
universal expressions, i.e. anger, disgust, fear, happy, sad,
and surprise along with seventh one neutral, are
recognized by the Generalized Feed-forward Neural
Network. Levenberg - Marquart (LM) nonlinear
optimization algorithm is used to train and test the Neural
Network. They have attained 93.3 % classification rate
with testing performance 0.0573.
A. R. Ardakany, in 2012 provided classification using a
simple feature extraction which performs geometric and
appearance features at the same time. This feature
extraction is carried out by computing the derivative in all
pixels of face images and then constructing a histogram
based on edges magnitudes and directions. The
experiments are clear evidence that presented method is
quite competitive with 95.67% accuracy on FERET
database [14].
A novel approach which is augmenting the classical model
with generic knowledge-based, WordNet was proposed by

Yohan in 2005. This approach attempts to trim irrelevant
keywords by the usage of Word Net. To categorize
irrelevant keywords, investigation on various semantic
similarities of keywords and to fuse the outcomes of all
these measures together to make a final decision with the
help of Dempster-Shafer evidence combination. Various
models have implemented by them to associate visual
tokens with keywords based on knowledge-based, Word
Net and evaluated performance using precision, and recall
using benchmark dataset. The results show that by
augmenting knowledge-based with classical model they
can improve precision of annotation by taking out
irrelevant keywords [15].
4. PROPOSED WORK
The proposed system design is given using figure 3 which
is given in two main phases. First for train the system or
storing the data into the database and second is used for
accepting the user query and producing the search results.
Therefore the entire system is described in three major
modules first feature extraction, Query interfacing and
finally the results listing.
Fig 3: Proposed system
4.1. Feature extraction
The content based images are retrieved by their image
properties such as image objects edges, color distributions
and the image textures. Therefore all the tree image
features are computed and normalized first which is
stored in a database table for image feature
representation. At the same time the image are also tagged
with some kind of text which indicates the objects
available in the input image during training phase. These
tags are preserved separately in a table. But in order to
recognize the image a key is assigned which is also
preserved with the databases.
4.2. Query interface
As database is filled with the image contents the training
session of the presented model is completed. Now for
accepting the user query the system can accept the text
query and image query also through the individual user
interface.
4.3. Search outcomes
The user produced query is supplied to the KNN algorithm
where the KNN having to inputs first the user query
tokens and second the database of images, image features
and the tag associated with images. Thus by finding the
distance between the user query input and the data base
scenarios the nearest distance images and their objects
are recognized.
The basic features of the proposed work model are
explained in this section. In addition of that their modular
distribution for implementation of the CBIR model is also
explained. The next section discussed the conclusion and
the future extension of the presented work.
5. CONCLUSIONS
n this age need of efficient and accurate systems are
increases due to uneven increasing demand of image
contents. Thus in this paper first a survey on Content
based image retrieval (CBIR) is performed in addition of
that the different approaches developed in recent years
are also discussed. During investigation that is found that
there are a number of techniques that have been
introduced so far but most of them are not much accurate
or sometimes lake in performance. This paper has
portrayed some of major contributions made in this field
and discusses the fundamentals. Additionally a new model
is also suggested for future implementation.
ACKNOWLEDGEMENT
This research paper is written by me with help of my PG
guide Mr. Nitesh Rastogi and HOD Mr. Anil Khandekar of
IIST, Indore. Thanks a lot of them.
REFERENCES
[1] Datta, Ritendra, Jia Li, and James Z. Wang. "Content-
based image retrieval: approaches and trends of the
new age." Proceedings of the 7th ACM SIGMM
international workshop on Multimedia information
retrieval. ACM, 2005.
[2] Paton, Norman W., and Oscar Díaz. "Active database
systems." ACM Computing Surveys (CSUR) 31.1
(1999): 63-103.
[3] Khokher, Amandeep, and Rajneesh Talwar. "Content-
based Image Retrieval: Feature Extraction

Techniques and Applications." International
Conference on Recent Advances and Future Trends in
Information Technology (iRAFIT2012). 2012.
[4] Madugunki, Meenakshi, et al. "Comparison of
different CBIR techniques." Electronics Computer
Technology (ICECT), 2011 3rd International
Conference on. Vol. 4. IEEE, 2011.
[5] Rui, Yong, et al. "Relevance feedback: a power tool for
interactive content-based image retrieval." Circuits
and Systems for Video Technology, IEEE Transactions
on 8.5 (1998): 644-655.
[6] Murthy, V. S. V. S., et al. "Content based image
retrieval using Hierarchical and K-means clustering
techniques." International Journal of Engineering
Science and Technology 2.3 (2010): 209-212.
[7] Siorpaes, Katharina, and Elena Simperl. "Human
intelligence in the process of semantic content
creation." World Wide Web 13.1-2 (2010): 33-59.
[8] Su, Ja-Hwung, et al. "Efficient relevance feedback for
content-based image retrieval by mining user
navigation patterns." Knowledge and Data
Engineering, IEEE Transactions on 23.3 (2011): 360-
372.
[9] Lai, Chih-Chin, and Ying-Chuan Chen. "A user-
oriented image retrieval system based on interactive
genetic algorithm." Instrumentation and
Measurement, IEEE Transactions on 60.10 (2011):
3318-3325.
[10] Ramesh BabuDurai, C., V. Balaji, and V. Duraisamy.
"Improved content based image retrieval using SMO
and SVM classification technique." European Journal
of Scientific Research 69.4 (2012): 560-564.
[11] Raghuwanshi, Ghanshyam, Nishchol Mishra, and
Sanjeev Sharma. "Content based image retrieval
using implicit and explicit feedback with interactive
genetic algorithm." International Journal of Computer
Applications 43.16 (2012): 8-14.
[12] Jayaprabha, P., and RmSomasundaram. "Content
Based Image Retrieval Methods Using Self Supporting
Retrieval Map Algorithm." IJCSNS 13.1 (2013): 141.
[13] Moore, S., and R. Bowden. "Local binary patterns for
multi-view facial expression recognition." Computer
Vision and Image Understanding 115.4 (2011): 541-
558.
[14] Ardakany, Abbas Roayaei, and A. M. Joula. "Gender
recognition based on edge histogram." International
Journal of Computer Theory and Engineering 4.2
(2012): 127.
[15] Jin, Yohan, et al. "Image annotations by combining
multiple evidence &wordnet." Proceedings of the
13th annual ACM international conference on
Multimedia. ACM, 2005.

A Survey on Image retrieval techniques with feature extraction

More Related Content

What's hot (18)

Similar to A Survey on Image retrieval techniques with feature extraction (20)

More from IRJET Journal (20)

Recently uploaded (20)

A Survey on Image retrieval techniques with feature extraction