2019 IEEE International Symposium on Smart Electronic Systems (iSES) (Formerly iNiS)

A Comprehensive Survey on Content Based

Image Processing Techniques
Anil Mishra
Tanmay Kasbe
Research Scholar, Computer Science Department
Oriental University Assistant Professor, Computer Science
Indore (M.P), India Department Oriental University
[email protected] Indore (M.P), India
[email protected]

Abstract— In modern computing world there is rapid growth

or indistinguishable shading histograms and pictures in use for
in image transaction. Most of the digital equipments generate
images for different operations. It creates huge repository of various encompassing lighting may create various histograms.
images. Accurate Retrieval of desired images from these Texture is one of the mainly significant, which planned to
repositories becomes a challenging task for every researcher. catch the granularity and redundant examples of surfaces
There are tremendous methods of retrieval of images. CBIR is inside a picture. At last the shape is a significant component
one of the well-liked techniques to pull out low level features of that is utilized for entity recognition. There are standard
images. Some of the features are Color, shape, texture and calculations for edge identification and some system could
Spatial Location. In the paper, we survey perception of CBIR, used to speak to and depict this shape be for further recovery
variety of techniques for retrieval, features of images, standard of comparable pictures.
image datasets. These features are taking out by using various methods. A
Keywords—Image Transaction, CBIR, Image Retrieval,
few of methods are conferred in this paper. The figure 1
Image Attributes displays the model of CBIR system. This CBIR system is
depends on the color, shape and texture characteristics. The
I. INTRODUCTION remaining paper is structured as follows. In Section I-A, we
discussed three main features of the system. Section I-B, lists
The use of images is increasing progressively by the
some of the benchmark dataset of the images with the brief
common user of the electronics devices. This causes the large
description. Section I-C, lists some of the software packages
amount of images in the database. The searching and retrieval
used to retrieve characteristics the pictures. While Section I-D,
of desire images are one the most requirements for the users
discuss few image Retrieval Methods. Section II describes the
from various areas like medicine, health, marketing,
literature review done by the researcher. They
computing, defense, fashion, design etc [1]. The recovering of
reviewed scholarly papers, which incorporate the existing
these pictures is a fascinating task. At whatever point a client
understanding including conclusions as well as offerings in
wishes to recover a single picture, multiple number of pictures
this field. Conclusions are talked in the Section III.
are recovered. The greater parts of them are irrelevant to
client’s request. If clients are framing arbitrary inquiries then
proficient and successful results can't be yielded. This makes
client unsatisfied more often. Part of endeavors has been done
towards proficient recovery of target picture from enormous
database has turned out to be significant. This presumably
inspires the researchers for broad examination of picture
recovery.
To access the precise and rapid desired images the Content
Based Image Retrieval (CBIR) is used. CBIR is a method that
uses low-level image properties. The majority of the
researchers apply features like color, shape and texture of
images. Color feature is frequently extensively used to portray
the pictures which are hard to be fragmented and needn't to
consider space data. Color is an emotional human vibe of
obvious light contingent upon force. HSV and RGB spaces
are utilized to signify Color. By and large, the conveyance of
Color was spoken to by shading histograms.

Shape and surface highlights were embraced to improve the

shading based CBIR as various pictures may have comparable
Fig. 1: Model of CBIR System

978-1-7281-4655-3/19/$31.00 ©2019 IEEE

DOI 10.1109/iSES47678.2019.00097 396
A. Image Features Table 1: Dataset Description

Color features Type of dataset Description

Wang Dataset Used in feature extraction

Color is the most popular visual feature of the image.
Human eyes are very receptive to color images than grey level Corel Dataset Thousand of images from various
images. RGB color space uses Red, Green and Blue color categories. Each category has 100
intensities. There are various color feature extraction methods images of JPEG format.
like color histogram, auto-correlogram, color moments etc. Contains only 20 categories. Each
GHIM Dataset
Color histogram is used to signify the color stuffing of the category has 500 images of JPEG
image. The image histogram displays the range of gray levels format.
0-255. To reduce this wide range of value, the color range is
divided into bins. Color moments compute the color allocation CIFAR-10 Dataset Contains images from 10 classes.
in an image. If the color allocation in the two images is same
then the two images are supposed to be a like. UW database There are 18 categories, for example
“spring flowers”
Color correlogram is a technique that evaluates the
It contains images of buildings from
distribution of color pair as a function of the space connecting ZuBuD database
different angles in different time of day
pixel pair. The method describes spatial correlation of duo of
and night.
colors alters with space. It calculates spatial correlation
between like colors. FIGRIM Dataset FIne-GRained Image Memorability
dataset contains 9428 images. The
images divided into 21 categories. Each
category contain 300 images of 700x700
Shape Feature pixels.

MIRFLICKR-25000 It consists of 25000 images from the

Shape feature of image, determine the edge determination social photography site Flickr website.
in an image. Shape feature categorized into two groups:
IRMA-10000 database It consists of 10,000 radio-graphs. Out
of which 9000 are training and 1000 are
a. Exterior border of the shape. test images.
b. Region of the shape.
To evaluate the above two categories Fourier descriptor and INRIA PERSON Dataset Dataset for finding of standing people.
Moment invariant methods are used.

Texture Feature Caltech Pedestrian Dataset It is a Pedestrian Detection

Benchmark. The training data have six
training sets and testing have five sets.
Texture feature of image, determine regular replication of a
component or pattern on a surface. Texture feature depicts the
structural collection of plane and their association with the
neighboring environment. The texture feature can be taken out C. Software Used
using GLCM (Grey level co-occurrence matrix), Wavelets,
Fourier transform, entropy, correlation methods. Gabor and Many software packages provide for feature extraction and
wavelet transform extracts statistical distribution of image. dimension reduction. Some of the packages like MATLAB,
The six texture properties are coarseness, contrast, roughness, SciLab, Julia the R language give user friendly environment
regularity, directionality and line likeness. for analyzing the image.
B. CBIR Benchmark Dataset
 Julia: Julia is a open source software.
 Scilab: The command syntax of SciLab and
Different types of dataset are available online to work with
MATLAB are very similar.
the imaging system. These dataset contain wide variety of
 R programming language: It is free environment
images from Tribal, Transport to Buildings. Some of the
software for graphical computing. It is a
datasets, we have illustrated in this section. Total eleven
command line interface.
datasets are discussed in this paper [22].
 MATLAB: Matlab is graphical user interface
developed by Mathwork. It is a short name of
Matrix Laboratory.

39
D. Image Retrieval Methods
Weighted harmonic mean of the
This section discusses the range of feature vector F-Measure
precision and recall.
calculation methods that are used in image retrieval system.
Hausdorff distance Distance between two subsets of
Color Feature Retrieval Methods (Grid Color Moment) a metric space.

The said method foremost changed the RGB image into Chessboard Distance Metric
HSV color space. Then this image is divided into three x computes the pathway between the
Chessboard Distance
three matrix blocks. Mean color, Variance and Skewness for pixels based on 8 attached nearby
H,S and V is computed for every block. Thus every block pixels.

has nine features and complete image have 81 features.

Distance between two points is the
Calculate mean and standard deviation for database. Finally Chamfer distance least of the related overheads to
perform the normalized feature value. all the paths from one point to
other.
Shape Feature Retrieval Methods (Canny Edge Detection)
Manhattan Distance It is the sum of the horizontal and
The method focuses on reducing the amount of data in the vertical sections.
image, without disturbing its structure. Smoothing,
Obtaining gradients, Non- greatest control, Double
threshold: and Edge tracking by hysteresis is the steps
involved in this method. II. LITERATURE SURVEY
U. P. Singh and S. Jain, [2] proposed method to measure the
Texture Feature Retrieval Methods (Local Binary Pattern) accuracy of image requests from different database images.
In this method LBP value for a given pixel is calculated The method considerably rectifies the feature extraction
with similarities of its adjoining. Following the identification procedure.
of LBP pattern of every pixel, a histogram is prepared to S. Patil and S. Talbar, [3] performed experiment and showed
build the texture photo. that the dominance of Canberra, Bray-Curtis, Square chord,
and Chi-squared distances over the conformist Euclidean and
E. Assessment Techniques Manhattan distances.
O.P. Yadav and V. Suryawanshi, [4] used Chessboard
The CBIR Structure is assessed on the basis of usefulness
and competence. These measures are related with the accuracy distance, Cityblock distance and Euclidean distance for
and speed of the image recovery. We are hereby discussing similarity matching. They used “Haar” wavelet
some of the popular techniques used in the system:
N. Singh, et al., [5] propose a framework to choose the most
Table 2: Description of Assessment Techniques suitable way to offer the quite appropriate features to
scrutinize real-time images.
Techniques Description M. Sharma & A. Batra, [6], performed experiment and
shows the outbalancing of Canberra, Bray-Curtis, Square
Euclidean Distance Straight-line distance between two chord, and Chi- squared gaps with respect to the
pixels. conservative Euclidean and Manhattan distances.

Chi Square Distance Similar to Euclidean distance, but it G. Duan , et al., [7], focuses on the feature withdrawal and
is weighted. demonstration, numerous frequently used algorithms and
picture matching methods.
Geodesic Distance Least path length between
two points on the surface. F. Malik and B. Baharudin, [8], proposed the strategy for
removing quantized histogram measurable surface from the
Bhattacharya Distance Calculate the comparability of DCT squares of the picture utilizing the huge vitality of the
two probability distributions.
DC and the initial three AC coefficients of the blocks.
Mahalanobis Distance Used to measure distance P. S. Hiremath and J. Pujari, [9], proposed a model to attain
between two groups. Used in
discriminant analysis. superior retrieval competence by picture and its match. The
picture and its match are divided into non-overlie block of
Precision is the %age of same dimension. The features drained from restrictive co-
Precision and Recall outcome which are pertinent, occurrence histograms between the image blocks and
Recall is the
%age of entire pertinent
analogous match block, in RGB color space, provide as local
outcome appropriately descriptors of color and texture. This data is taken for 2
classified. resolutions and 2 grid designs that give diverse facts of the
like picture.

39
J. Yue, et al., [10] designed a CBIR system with color and highlights. The traditional Gabor change has a few
texture merged properties by building weights of feature restrictions. The symmetrical positions are not framed by
vectors. The considerable recovery try outs depict that the Gabor work that is the reason its portrayal isn't strong and it
merged properties recovery takes superior illustration effect expends more calculation cycles and more memory.
than the solo feature recovery, gives good consequences.
Peizhong Liu et al., [17] used Color Information Feature
R. K Lingadalli1 and N.Ramesh, [11], proposed a (CIF), and the LBP-based feature in the photo recovery and
calculation which consolidates every one of the three categorization schemes. The hybrid CIF and LBP-based
highlights, for example, Color, shape and texture. The attribute presents a promising outcome and outflanks the
exactness of HSV color space depends on color histogram current strategies more than a few picture databases.
based liking result better outcome. The rate of shape based
retrieval can be enhanced by considering estimated shape in M. A. Ansari, et al., [18], built up the two stage procedure
spite of the require shape. Grey Level Co-occurrence matrix where the initial stage is to remove the low level highlights
(GLCM) is used to take out the texture properties of the of the picture by utilizing shading, surface and shape
photos. The feature identical procedure depends on the descriptor. (CEDD) for removing the shading just as surface
Canberra distance. highlights and 2-stage distinct wavelet change (2D-DWT)
for extricating the shape include. In the later stage the SVM
R. Chaudhari and A. M. Patil, [12], anticipated an algorithm classifier is utilized for arranging the pictures in various
to make perfect the precision and act of recovery. The classes.
precision of color histogram based identical can be increased
with Color Coherence Vector (CCV) for following fine- W. Jiji and P. Johnson Durai Raj, [19], proposed method to
tuning. The rate of shape based recovery can be improved concentrate shading and shape highlights for the examination
via estimated shape in spite of the precise shape. of dermatology pictures. They proposed calculation for
shading and shape include vectors and the highlights are
Y. Rao , P. Mundur and Y. Yesha [13] and X. L. X. W. D.C. standardized utilizing Min–Max standardization. Molecule
Tao and X. Tang [14] uses SVM to pull out the image features swarm enhancement (PSO) method for numerous category
precisely and recover the required image competently. A large characterizations is utilized to combine the hunt space all the
portion of the work done around there has utilized SVM more proficiently.
(Support Vector Machine) for the pattern categorization. The
learning technique is little and straight-forward [15]. SVM is A. Saini and S. Singh, [20], assess the presentation of CBIR
supposed as a tough aspirant for categorization as of framework with parameters accuracy, review, NMRR and
numerous causes: its generality capability, speedy learning, recovery time by utilizing highlight extraction methods
suppleness and estimation of consequence feedback. dependent on shading - histogram, surface - GLCM and area
– limit descriptors of a picture. NMRR and recovery time of
M. Kokare et al., [16] has demonstrated that the exhibition various question pictures have been assessed.
and proficiency of Gabor wavelet do superior when
contrasted with the ordinary symmetrical wavelet based
Table 3: Techniques used in various research

S.No. Author and Reference Method Dataset Contribution Findings

Co-occurrence
histograms The outcomes are better
Gradient Vector Flow Proposed a model for attaining than SIMPLIcity,
P. S. Hiremath and J. Pujari
1 fields Corel Dataset superior recovery competence FIRM, edge based and
[9]
Invariant Moments by photo and its match. Histogram based
integrated image system.
matching
The projected technique
provides superior
recovery outcomes and
HSV color space Suggests algorithms for the
average exactness. The
R.K Lingadalli and N.Ramesh GLCM Archaeological images and betterment of precision and
2 technique uses different
[11] Canberra distance Wang’s dataset performance of recovery of
GLCM statistical
measure pictures.
properties with angle
(0,45,90,130) and
distance d=1.

Used to analyze dermatology

The technique applied
images. The algorithm
on total 1450 photos.
G. W. Jiji and P. J. Durai Raj Min–Max normalization performs efficient recovery of
3 Webite medicine net This gives 98.22%
[19] PSO technique pictures of skin lesions and it
helps greatly in finding of skin specificity and 94%
lesions automatically. sensitivity.

39
 The method applied on
diverse pictures dataset
The approach consists of 2 and show that it is
CEDD foot-steps. Initially take out offering improved
Wang image database outcomes over present
2D-DWT the low level characteristics
4 M. A. Ansari, et al., [18], Caltech image database techniques like CEDD +
SVM Classifier and then categorize the
Corel image database SVM and 2D DWT +
Euclidean distance pictures into different
categories.. SVM in terms of
Precision Recall and
Accuracy.

Edge Histogram
Descriptor
Color Auto Correlogram The method consolidates basic
IMAGEVARY Improved precision and
5 P. Vikhar, and P. Karde [23] Color Moments features and then uses SVM
recall value.
Gabor Wavelet classifier.
Transform
SVM
Corel Dataset Combined CIF and LBP-
Outex TC00000 Used CIF and LBP in the based attribute provides
CIF
6 P. Liu et al., [17]
LBP
TC00001 photo recovery and a hopeful outcome that
Brodatz categorization systems. outflanks the current
USPTex strategies.
The significant recovery
tries out depicting that
HSV color space the merged properties
J. Yue et al., [10] Build a CBIR system using
7 Color histogram SQL server 2005 database shows superior
weights of feature vectors.
illustration effect in spite
of solo feature, gives
good consequences.

2012.
III. CONCLUSION
[6] Meenakshi Sharma & Anjali Batra, “Analysis of Distance
The paper reviewed the basic CBIR model for image Measures in Content Based Image Retrieval”, Global Journal of
feature retrieval. Here we have discussed three different basic Computer Science and Technology, Vol. 14 Issue 2, 2014
features, different types of benchmark database with their [7] Guoyong Duan, Jing Yang and Yilong Yang, “Content-Based
features, software used to extract the features and the Image Retrieval Research”, . SciVerse ScienceDirect, Procedia
techniques used in the calculation of feature extraction. The Technology, Physics Procedia Vol. 22, pp. 471 – 477, 2011.
algorithm tries to recover speedy and accurate calculation of [8] Fazal Malik , Baharum Baharudin, “Analysis of distance metrics
the features. In addition to these fast and high speed gadgets in content-based image retrieval using statistical quantized
will be developed to support these algorithm. histogram texture features in the DCT domain”, Journal of King
Saud University – Computer and Information Sciences Vol. 25,
pp. 207– 218, 2013.
[9] P. S. Hiremath and Jagadeesh Pujari, “Content Based Image
REFERENCES
Retrieval based on Color, Texture and Shape features using
[1] Swati Sonawane and V. S. Patil, “Image Retrieval Based On Image and its complement”, International Journal of Computer
Texture- A Survey”, International Journal of Industrial Science and Security, Vol. 1, Issue 4, 2007.
Electronics and Electrical Engineering, Vol. 4, Issue-7, 2016. [10] Jun Yue, Zhenbo Li, Lu Liu and Zetian Fu, “Content- based
[2] Uday Pratap Singh and Sanjeev Jain, “Content Based Image image retrieval using color and texture fused features”, Elsevier
Retrieval Using Euclidean and Manhattan Metrics”, Journal of Ltd. SciVerse ScienceDirect, Mathematical and Computer
Mathematical Sciences: Advances and Applications, Vol. 4, No. Modelling Vol.54, pp. 1121–1127, 2011.
1, pp 217-226, 2010. [11] Ramesh K Lingadalli1 and N.Ramesh, “Content Based Image
[3] Sanjay Patil and Sanjay Talbar, “Content Based Image Retrieval Retrieval using Color, Shape and Texture”, International
Using Various Distance Metrics”, R. Kannan and F. Andres Advanced Research Journal in Science, Engineering and
(Eds.): ICDEM 2010, LNCS 6411, pp. 154–161, 2012. Technology Vol. 2, Issue 6, 2015.

[4] Omprakash Yadav and Vaishali Suryawanshi, “CBIR [12] Reshma Chaudhari and A. M. Patil, “Content Based Image
Retrieval Using Color and Shape Features”, International
Evaluation using Different Distances and DWT”, International
Journal of Advanced Research in Electrical, Electronics and
Journal of Computer Applications (0975 – 8887) Vol. 93, No
16, 2014. Instrumentation Engineering Vol. 1, Issue 5, 2012.

[5] Nidhi Singh, Kanchan Singhb and Ashok K. Sinha, “A Novel

Approach for Content Based Image Retrieval”, Elsevier Ltd.
SciVerse ScienceDirect, Procedia Technology 4 pp. 245 – 250,

40
[13] Y. Rao , P. Mundur and Y. Yesha, “Fuzzy svm ensembles for
relevance feedback in image retrieval”, LNCS, vol. 4071, pp.
350-359, 2006.

[14] X. L. X. W. D.C. Tao and X. Tang, “Asymmetric bagging and

random subspace for svm-based relevance feedback in image
retrieval”, IEEE Transactions on Pattern Analysis and Machine
Vision, Vol. 28 No. 7, pp.1088-1099, 2006.

[15] H.-Y. Y. Xiang-Yang Wang, Jing-Wei Chen, “A new integrated

SVM classifiers for relevance feedback content-based image
retrieval using em parameter estimation,” Applied Soft
Computing, Vol. 11, pp. 2787- 2804, 2011.

[16] M. Kokare, P.K. Biswas and B.N. Chatterji, “Texture image

retrieval using rotated wavelet filters”, Pattern Recognition
Letters, Vol. 28, p. 1240-1249, 2007.
[17] Peizhong Liu , Jing-Ming Guo , Kosin Chamnongthai , Heri
Prasetyo , “Fusion of Color Histogram and LBP- based
Features for Texture Image Retrieval and Classification”,
Information Sciences, 2017.
[18] Mohd. Aquib Ansari, Manish Dixit, Diksha Kurchaniya and
Punit Kumar Johari, “An Effective Approach to an Image
Retrieval using SVM Classifier”, International Journal of
Computer Sciences and Engineering, Vol. 5, Issue 6, pp. 2347-
2693, 2017.

[19] G. Wiselin Jiji and P. Johnson DuraiRaj, “Content- based image

retrieval techniques for the analysis of dermatological lesions
using particle swarm optimization technique”, Elsevier,
ScientDirect, Applied Soft Computing Vol. 30, pp. 650–662,
2015.

[20] Aman Saini and Sonit Singh, “Evaluation of CBIR System by

Using Low Level Features of an Image”, International Journal
of Science, Engineering and Technology Research (IJSETR)
Vol. 5, Issue 6, 2016

[21] S.Rubini, R.Divya, G.Divyalakshmi and T.M. Senthil Ganesan,

“Content Based Image Retrieval (CBIR)”, International
Research Journal of Engineering and Technology (IRJET),
Vol. 5, Issue 3, 2018.

[22] Ankitha Varma and Kamalpreet Kaur, ““Survey on content

based image retrieval” “,International Journal of Engineering &
Technology, Vol. 7, No. 4.5, pp. 471-476, 2018.

[23] Pradnya Vikhar, and Pravin Karde, “Improved CBIR

System using Edge Histogram Descriptor (EHD) and
Support Vector Machine (SVM)”, International Conference
on ICT in Business Industry & Government” IEEE, Vol.
978, No. 1, pp. 5090-5515, 2016.

[24] Neha Ghosh, Shikha Agrawal and Mahesh Motwani,” A

Survey of Feature Extraction for Content-Based Image
Retrieval System”, Proceedings of International
Conference on Recent Advancement on Computer and
Communication, Springer Nature Singapore Pte Ltd. Vol.
1007, No. 10, pp. 978-981, 2018.

40