




































 

 

          ISSN : 2693 6356 

     2022 | Vol 5 | Issue 6 

 

 

Web Image Search Engine with Integrated Feature Extraction 
Dr. K. Gayatri 

1
, Dr. K. Sandeep  Raja

2
 

1,2
Professor, Department of Computer Science and Engineering 

1,2
Malla Reddy Institute of Engineering and Technology, Hyderabad, Telangana, India. 

 

 

 

Abstract 
The spatial dependence matrix (SDM) is used to extract the texture features from the provided 

images, and the translation invariant discrete wavelet transform (TIDWT) is used for low-level 

feature extraction, making for an effective content-based image retrieval (CBIR) system. To 

further demonstrate the efficacy of the suggested hybrid CBIR system, this method also took into 

account the Tanimoto distance. Extensive experimental research shows that the efficiency of the 

proposed hybrid CBIR system is much higher than that of traditional CBIR systems. 
Key words: Case-based information retrieval, Spatial Dependence Matrices, Discrete 

Wavelet Transforms, Measuring Similarity, and Accuracy 

 

 

 

 
1. Introduction 

In recent years, digital information 

acquisition has gained popularity as a result 

of the fast growth of computer technology. 

Multiple gigabytes of data including 

photographs are produced daily. Some 

examples include the armed forces, 

medicine, and the media. Because of the 

prevalence of digital material, efficient 

methods of storing and retrieving visual 

data are urgently required. Textual queries 

and content-based queries are the two most 

used methods for retrieving photos. Text-

based methods are quite popular and 

frequently utilized. A user may use this 

method to find a certain picture by 

searching for it using keywords. Despite its 

ease of use, the system has limitations due 

to the reliance on human perception for the 

annotation of images. This indicates that the 

picture search terms used depend on the 

language being used. Content Based Image 

Retrieval (CBIR) [1] is an alternative to 

traditional text-based methods. CBIR is a 

method for retrieving pictures from a 

database according to a user's query based 

on the image's visual characteristics [2]. In 

addition to its use in crime prevention, the 

CBIR system has found usage in the fields 

of architecture, engineering, fashion, 

journalism, advertising, medical diagnosis, 

home entertainment, web searching, and 

many more. 

Below is the rest of the paper: In Section 2, 

we will discuss the work that has been 

accomplished so far in the CBIR area and 

the limitations of the current DWT-based 

CBIR system. Section 3 describes the 

proposed CBIR methodology, including 

information on TIDWT and SDM 

techniques. In Section 4, we explore the 

findings and their implications. Section 5 

has the conclusions, then the bibliography. 

 
2. Related Work 

In the past decades several CBIR 

systems have been proposed, and still 

the researchers are focusing on 

developing extended CBIR systems 

with more effective results. The letter 

proposed in [3] gives a comparison of 

different approaches of CBIR based 

on similarity measures and image 

features to identify the similarity 

between the images, which provides 

accurate information for retrieving 

the relevant images from large 

database. Wan Siti et.al proposed in 

[4] compares the several medical 

image retrieval systems based on the 

feature 

extraction and to improve the 

effectiveness of the CBIR system for 

medical images such as magnetic 



 

 

resonance (MR) images and 

computed tomography (CT) images 

[9]. The major concept proposed in 

[4] is to help in the diagnosis such as 

to find the similar disease and 

monitoring of patient s progress 

continuously. B. S. Manjunath et.al 

presented in [5] is the combination of 

color, texture with inclusion of edge 

compactness for Motion Picture 

Expert Group (MPEG)-7 standards. 

Another approach proposed in [6] 

used different color spaces such as 

HSV and YCbCr explains a similar 

approach based on color and texture 

analysis. The work proposed in [7] 

introduces a new retrieval system 

which has done by using wavelet 

transformation with both color and 

texture features together and will 

perform better than existed state of art 

algorithms. Retinal image retrieval 

system called CBIR for retinal and 

blood vessels extraction is presented 

in [8] which has been analyzed by the 

histogram features of RGB color 

components. The multi resolution 

analysis has applied to the image to 

acquire the texture information. In 

addition to improve the performance, 

morphological operations are applied 

to study the shape of object. 

Swati Agarwal has proposed a 

new CBIR system in [10], which is 

by using discrete wavelet transform 

and edge histogram descriptor 

(EHD). Here the retrieval is based on 

color and texture features not by 

using color information in the image, 

input image first decomposes the 

input query image into several sub 

bands i.e., approximation coefficients 

and detail coefficients, where detail 

coefficients consists of horizontal 

(LH), vertical (HL) and also the 

diagonal information (HH) of the 

image. Afterwards, EHD is used to 

gather the information of dominant 

edge orientations. This mixture of 

3D-DWT and EHD will improve the 

efficiency of the CBIR system. 

Recent times, many researchers 

focused on wavelet based CBIR 

system by adopting any texture 

feature extraction approach [11-14]. 

However, DWT suffers from lack of 

translation invariancy which plays a 

key role in extracting low-level 

features from input image while 

decomposing it. In latter subsection, 

the process of DWT and its 

drawbacks explained in detail. 

2.1. Discrete Wavelet Transform 

Discrete Wavelet Transform 

(DWT) is a modified version of 

Continuous Wavelet Transform 

(CWT). DWT principles are very 

similar to the CWT however the 

wavelet scales and positions are based 

upon powers of two. The basic 

principle of DWT is to pass the input 

signal through a group of filters i.e., 

low pass and high pass filters to get 

the low frequency (LF) and high 

frequency (HF) of source signal. Low 

frequency contents include LL and 

these coefficients are known as the 

approximation coefficients. This 

means the approximations are 

obtained by using the high scale 

wavelets which corresponds to the 

low frequency. The high frequency 

components which are known as LH, 

HL and HH of the signal are called 

the details which will be obtained by 

using the low scale wavelets which 

corresponds to the high frequency. 

The process of DWT filtering 

includes, first the signal is fed into the 

wavelet filters. These wavelet filters 

comprise of both the high-pass and 

low- pass filter. Then, these filters 

will separate the high frequency 

content and low frequency content of 

the signal. However, with DWT the 

numbers of samples are reduced 

according to dyadic scale. This 

process is called the sub-sampling. 

Sub-sampling means reducing the 

samples by a given factor. Due to the 

disadvantages imposed by CWT 

which requires high processing power 

the DWT is chosen due its simplicity 

and ease of operation in handling 

complex signals. 

However, this approach has been 

suffering from the decimation 

property of DWT, because when we 

apply the DWT decomposition to an 

image, it will decompose the image 

into four sub bands with reducing the 

size of it to the half of actual image 

size. Due to this decimation we lose 

some original information while 

processing with DWT. Therefore, to 

improve the performance of CBIR 

system further, one needs to gain the 



 

 

lost information and add it with high 

frequency sub bands to get efficient 

features at low-level. 

3. Proposed System 

3.1. Translation Invariant Discrete 

Wavelet Transform 

It is a wavelet transform algorithm 

designed to overcome the lack of 

translation- invariance of DWT. 

Translation-invariance is achieved by 

removing the down samplers and up 

samplers in the DWT and up 

sampling the filter coefficients by a 

factor of 2(𝑗−1) in the 𝑗𝑡ℎ level of the 

algorithm. The TIDWT is an 

inherently redundant scheme as the 

output of each level of TIDWT 

contains the same number of samples 

as the input – so for a decomposition 

of 𝑁 levels there is a redundancy of 

𝑁 in the wavelet coefficients. The 

following block diagram depicts the 

digital implementation of TIDWT. 
 

 

 
Figure 1. TIDWT filter bank with 3 
levels 

 

In the above diagram, filters in each 

level are up-sampled versions of the 

previous (see figure below). 

 

 
Figure 2. TIDWT filters 

 

3.2. Spatial Dependence Matrix 

It is known as the texture 

investigated method with statistics in 

which the pixels spatial relationship 

is considered in a range of greyscale. 

Texture characterization of a source 

image is done by computing how 

frequently pixel pairs with specific 

intensity values and the occurrence of 

specified spatial relationship in an 

image. Then after, computing the 

extracted measure of statistics from 

this obtained SDM matrix provides 

texture information of source image. 

3.2.1. Contrast 

It gives the intensity contrast 

measurement of a pixel to its 

neighbour over the whole word 

image 

𝐶𝑜𝑛𝑡𝑟𝑎𝑠𝑡 = ∑𝑖,𝑗|𝑖 − 𝑗|2𝑝(𝑖, 

𝑗)                                                                      
(1) 

Where 𝑝 is a word image with a size of 𝑖 
× 𝑗 in which the number of rows denoted 

by 𝑖 and number of columns denoted by 𝑗 

3.2.2. Correlation 

It returns a measure of how correlated 

a pixel to its neighbour over the whole 

word image with a range of -1 to 1 

𝐶𝑜𝑟𝑟𝑒𝑙𝑎𝑡𝑖𝑜𝑛 = ∑ 
(𝑖−𝜇𝑖)(𝑗−𝜇𝑗)𝑝(𝑖,𝑗) 

 

(2) 
𝑖,𝑗 𝜎𝑖𝜎𝑗 

𝜇 represents the mean value and 𝜎𝑖, 𝜎𝑗are 
the variances of 𝑖and 𝑗 

3.2.3. Energy 

It provides the squared sum of the 

elements presented in gray scale matrix 

of word image 

𝐸𝑛𝑒𝑟𝑔𝑦 = ∑𝑖,𝑗 𝑝(𝑖, 𝑗)2 

3.2.4. Homogeneity 

(3) 

It will be used to measure the 

closeness of the elements distribution in 

the gray scale matrix to its diagonal 



 

 

𝑖,𝑗 
𝐻𝑜𝑚𝑜𝑔𝑒𝑛𝑖𝑒𝑡𝑦 = 

∑ 
𝑝(𝑖,𝑗)

 
1+|𝑖−𝑗|(4) 

 

 

Where 𝑝 is a word image with a size of 𝑖 
× 𝑗 

 

4. Simulation Results 

This section describes the 

simulation results of CBIR system 

based on the proposed and 

conventional schemes like DWT, 

DWT-EHD and DWT-EHD-HSV. It 

is implemented on MATLAB 2016b 

on a PC with Intel dual core 3
rd

 

generation processor having 8 GB of 

RAM capacity. The database consists 

of 1000 images belonging to 10 

different classes. There are 100 

images from each category. The 

images are of size 256x256 pixels. 

Out of these 100 images from each 

class, 25 images are used as query, 

while rest 75 images serve as database 

from which the similar images are 

retrieved. Thus, in all 75 images from 

each class are tested against 25 

different queries and the performance 

of the algorithm is analyzed. The top 

12 retrieved images of proposed 

CBIR system is disclosed in Figure 3, 

Figure 4 and Figure 5 respectively. 

 

Figure 3. Flower as a query and the top 12 retrieved images from 
database 

 

Figure 4. Elephant as a query and top 12 retrieved images from database 



 

 

 

Query image 

 

Figure 5. Dinosaur as a query and top 12 retrieved images from database 
 

5. Conclusion 

This article proposed an efficient 

hybrid CBIR system with the integration 

of SDM and TIDW transform which 

extracted both texture and low-level 

features. Further, Tanimoto 

distance is utilized for similarity 

measurement. Extensive experimental 

analysis disclosed that proposed hybrid 

CBIR system performed superior to the 

conventional CBIR systems. 

 
References 

[1] D. Feng, W. C Siu, and H.J. Zhang, 

“Fundamentals of Content Based Image 

Retrieval, in Multimedia Information 

Retrieval and Management 

Technological Fundamentals and 

Applications”, New York: Springer, 

2003. 

[2] A. W. M. Smeulders, M. Worring, S. 

Santini, A. Gupta, and R. Jain, 

“Content-based image retrieval at the 

end of the early years,” IEEE 

Transactions on Pattern Analysis and 

Machine Intelligence, Vol. 22, pp. 

1349–1380, Dec-2000. 

[3] Khan, S.M.H., Hussain, A. ; Alshaikhi, 

I.F.T., “Comparative Study on Content 

Based Image (CBIR),” International 

Conference in Advanced Computer 

Science Applications and Technologies 

(ACSAT), 2012. 

[4] Wan Siti, H. Munirah, W. Ahmad, M. 

Faizal and A. Fauzi, “Comparison of 

Different Feature Extraction Techniques 

in Content Based Image Retrieval For 

CT Brain Images,” 10Th IEEE 

workshop on Multimedia Signal 

Processing, pp. 503-508, 2008. 

[5] B. S. Manjunath, J. R. Ohm, V. V. 

Vasudevan, A.Yamada, “Colour and 

Texture Descriptors,” IEEE 

Transactions on Circuits and Systems 

for Video Technology, 1998. 

[6] P. S. Hiremath, S. Shivashankar, J. Pujari, 

“Wavelet Based Features for Color 

Texture classification with Application 

to CBIR,” International Journal of 

Computer Science and Network 

Security, Vol. 6, No.9A, September 

2006. 

[7] Tian Yumin, Mei Lixia, “Image Retrieval 

Based on Multiple Features Using 

Wavelet,” 5th IEEE International 

Conference on Computational 

Intelligence and Multimedia 

Applications (ICCIMA’03), 2003. 

[8] M. R. Zare, R. N. Ainon, W. C. Seng, 

“Content-based Image Retrieval for 

Blood Cells,” Third Asia International 

Conference on Modeling & Simulation, 

2009. 

[9] Prasad, B.G., Krishna, A.N., “Statistical 

Texture Feature Based Retrieval and 

Performance Evaluation of CT Brain 

Images” 3rd International Conference on 

Electronics Computer Technology 

(ICECT), Vol. 2, April 2011. 

[10] Swati Agarwal, A.K. Verma, Preethvanti 



 

 

Singh, “Content Based Image Retrieval 

using Discrete Wavelet Transform and 

Edge Histogram Descriptor” 

International conference on Information 

Systems and Computer Networks, 2013. 

[11] M. A. Ansari and M. Dixit, “An 

enhanced CBIR using HSV 

quantization, discrete wavelet transform 

and edge histogram descriptor”, 

International Conference on Computing, 

Communication and Automation, IEEE, 

Greater Noida, India, May 2017. 

[12] D. R. Dhotre, G. R. Bamnote and P. H. 

Shekokar, “Multilevel Haar wavelet 

transform and histogram based relevant 

image retrieval system”, International 

Conference on Computing 

Methodologies and Communication, 

IEEE, Erode, India, July 2017. 

[13] A. Nazir, R. Ashraf, T. Hamdani and N. 

Ali, “Content based image retrieval 

system by using HSV color histogram, 

discrete wavelet transform and edge 

histogram descriptor”, International 

Conference on Computing, Mathematics 

and Engineering Technologies, IEEE, 

Sukkur, Pakistan, Mar. 2018. 

[14] F. Nausheen, R. Kamble and M. Kokare, 

“Image Retrieval based on Wavelet and 

Optimized Local Gaussian Difference 

Extrema Pattern”, IEEE 13th 

International Conference on Industrial 

and Information Systems, Rupnagar, 

India, Dec. 2018. 


