VISVESVARAYA TECHNOLOGICAL UNIVERSITY BELAGAVI ...

VISVESVARAYA TECHNOLOGICAL UNIVERSITY

BELAGAVI-590010

A PROJECT REPORT ON

“SEARCHING AND RETRIEVAL OF IMAGES USING

DESCRIPTORS AND DISTANCE MEASURE”

Submitted in partial fulfillment of the requirements for the award of the degree of

Master of Technology

In

Software Engineering

2014-15

Submitted by:

SUBATHRA MUTHURAMAN (1NH13SSE17)

Under the Guidance of

Mrs.Swathi Baswaraju Asst. Prof, Dept.of ISE, NHCE

Department of Information Science and Engineering

New Horizon College of Engineering

Outer ring road, Kadubeesanahalli, Near Marathahalli, Bengaluru-590010

NEW HORIZON COLLEGE OF ENGINEERING(ISO-9001:2000 certified, Accredited by NBA,

Permanently affiliated to VTU)Outer ring road, Kadub eesanahalli, Near Marathahalli, B engaluru-s 60 1 03

DEPARTMENT OF INFORMATION SCIENCE ANDENGINEERING

NHCE,- | ru punsurr or exceuruqE l5r.

CERTIFICATE

Certified that the project work entitled "SEARCHING AND RETRIEVAL OFIMAGES USING DESCRIPTORS AND DISTANCE MEASURE" is a bonafide

work carried out by SUBATHRA MUTHURAMAN, with USN:1NH13SSEL7, rn

partial fulfillment for the award of Master Of Technology in Software Engineering ofthe Visvesvarilya Technological University, Belagavi during the year 2014-2015. It rscertified that all corrections/suggestions indicated for internal assessment have been

incorporated in the repoft deposited in the department library. The project report has

been approved as it satisfies the academic requirements in respect of Project workprescribed for the said degree.

\\9"r-Sighature of Guide

(Mrs.Swathi Baswaraju)

\L<*",SignatufofuOO

(Dr.Ajeet A.C) (Dr.Manjunatha)

External Viva

Name of Examiner

1.

2.

Signature with date

DECLARATION

I, SUBATHRA UIUTHURAMAN (USN:1NH13SSE17) , student of 4th semester

M.tech in Software Engineering, New Horizon College of Engineering, Bengaluru, hereby

declare that the project entitled "searching and Retrieval of images using Descriptors and

Distance measure" submitted to the Visvesvaraya Technological University during the

academic year 2014-2015,is a record of an original work done by me under the guidance ofMrs.Swathi Baswaraju, Assistant Professor, Department of Information Science and

Engineering, New Horizon College of Engineering, Bengaluru. This project work issubmitted in partial fulfilment of the requirements f,or the award of the degree of Master ofTechnology in Software Engineering. The results embodied in this thesis have not been

submitted to any other university or institute for the award of any degree.

"Cuirr*f"o [-fuffi rcrn4ct]^,

Date: O.j,,95" Jot.f

Place: Bengaluru

SUBATHRA MUTHURAMAN

(1NH13SSE17)

ACKNOWLEDGEMENT

Any project is a task of great enormity and it cannot be accomplished by an individual

without support and guidance. I am grateful to a number of individuals whose professional

guidance and encouragement has made this project completion a reality.

I have a great pleasure in expressing my deep sense of gratitude to the Chairman

Mr. Mohan Manghnani for having provided me with a great infrastructure and well

furnished labs.

I take this opportunity to express my profound gratitude to the Principal

Dr.Manjunatha, New Horizon College of engineering, Bengaluru for his constant support

and management.

I am grateful to Dr.Ajeet A.C, Professor and Head of Department, Department of

Information Science and Engineering, New Horizon College of Engineering, Bengaluru for

his strong enforcement on perfection and quality during the course of my project work.

I would like to express my thanks to Mrs.Swathi Baswaraju, Assistant Professor,

Department of Information Science and Engineering, New Horizon College of

Engineering, Bengaluru who has always guided me in detailed technical aspects throughout

my project.

I would like to express my thanks to the coordinators Mr.Lokesh M.R,Senior

Assistant Professor, Department of Information Science and Engineering and

Ms.Nandhini.N , Jr.Assistant Professor, Department of Information Science and

Engineering, New Horizon College of Engineering, Bengaluru for their valuable suggestions

throughout my project.

I would like to mention special thanks to all the Teaching and Non-Teaching staff

members of Information Science and Engineering department, New Horizon College of

Engineering, Bengaluru for their invaluable support and guidance.

Subathra Muthuraman (1NH13SSE17)

ABSTRACT

With explosive growth in social and multimedia, huge quantities of images are

available in web. These images are very good source of information in order to collect data

and make good use of it, if there is a good image retrieval system. The images that are

searched and retrieved are available with different visual and semantic content. The proposed

system targets to cater the growing need for an efficient image searching and retrieval system

for social/web media industry. The proposed system aims at using spatial context information

along with local features to improve the accuracy of searching and retrieving near matching

images. Spatial context information relates the object of interest with surrounding objects.

This spatial context information is converted in to binary codes which are used to perform

geometric verification between images. In the proposed method, interest points are identified

in the image using Hessian affine detector. Speeded up Robust Feature and Scale Invariant

Feature Transform descriptors are used to detect, describe the interest points in the image in

order to improve the accuracy of the retrieval system.

i

Contents

Page no.s

Abstract i

Contents ii

List of Figures iv

List of Tables v

List of Graphs vi

Chapters Titles Page no.s

Chapter 1 INTRODUCTION 1-4

1.1 Image Retrieval 1

1.2 Existing system 2

1.3 Objective of the project 2

1.4 Proposed system 2

1.5 Applications of CBIR 3

1.6 Dissertation Organisation

3

Chapter 2 LITERATURE SURVEY ON IMAGE RETRIEVAL 5-11

2.1 Literature Review

5

Chapter 3 ARCHITECTURE OF CBIR SYSTEM 12-17

3.1 Requirement Analysis and Specification 12

3.2 Architecture of the proposed system 15

3.3 Workflow of the system

16

Chapter 4 LOW LEVEL DESIGN 18-21

4.1 Class diagram 18

4.2 Use-case Diagram 19

4.3 Sequence Diagram

ii

20

Chapters Titles Page no.s

Chapter 5 IMPLEMENTATION 22-27

5.1 Tool used for implementation 22

5.2 Platform used for implementation 22

5.3 Module Implementation 23

5.3.1 Pre Processing 23

5.3.2 Query Processing Module 23

5.3.3 Feature Similarity Matching and Retrieval

27

Chapter 6 TESTING 28-30

6.1 Unit Testing 28

6.2 Integration Testing 29

6.3 System Testing

30

Chapter 7

Chapter 8

RESULTS AND DISCUSSION

CONCLUSION AND FUTURE WORK

31-38

39

REFERENCES

APPENDIX A

APPENDIX B

APPENDIX C

40-42

43-44

45-53

54-70

iii

List of Figures Page no.s

Fig 1.1 Concept of CBIR system 3

Fig 2.1 Approaches in CBIR system 5

Fig 3.1 Architecture of the proposed CBIR system 15

Fig 3.2 Work Flow Diagram of CBIR system 16

Fig 4.1 Class Diagram 18

Fig 4.2 Use-Case Diagram 19

Fig 4.3 Sequence Diagram 20

Fig 5.1 Scale Space Extrema Detection 24

Fig 5.2 Orientation Assignment 25

Fig 5.3 Keypoint Descriptor 26

Fig 7.1 Dataset used for the proposed system 31

Fig 7.2 Different categories of images used for the proposed system 32

Fig 7.3 Retrieved list of images 36

Fig B.1 Main Screen 45

Fig B.2 Dataset images 46

Fig B.3 Detected interest points 47

Fig B.4 Combined features 48

Fig B.5 Hessian detector points 49

Fig B.6 SIFT descriptor points 50

Fig B.7 SURF descriptor points 51

Fig B.8 Spatial context information-Hessian points 52

Fig B.9 Spatial context information-SIFT points 52

Fig B.10 Spatial context information-SURF points 53

Fig B.11 Result display 53

List of Tables Page no.s

Table 6.1 Unit Testing Table 28

Table 6.2 Integration Testing Table 29

Table 6.3 System Testing Table 30

Table 7.1 Precision and Recall for each category 35

Table 7.2 Mean Average Precision and Recall 37

v

List of Graphs Page no.s

Graph 7.1 Precision and Recall for each category 35

Graph 7.2 Mean Average Precision and Recall 37

vi

Searching and Retrieval of Images using Descriptors and Distance Measure

Dept. of ISE, NHCE 2014-2015 Page 1

CHAPTER 1

INTRODUCTION

With explosive growth in social and multimedia, huge quantities of images are

available in web. These images are very good source of information in order to collect data

and make good use of it, if there is good image retrieval system. The images that are searched

and retrieved are available with different visual and semantic content. This thesis is focussed

on developing efficient methods to find similar images which will be used in many fields like

social media, medical field, etc.

1.1 Image Retrieval

Image retrieval refers to finding and retrieving images from a large database of digital

images. Development of efficient image retrieval techniques would increase the utilization of

such data. Presently, images are retrieved by using Text descriptors and only a few depends on

Content Based Image Retrieval (CBIR), which searches and retrieves digital images from a

huge database, where content, refers to the color, texture and shape of the images [20].

Content Based Image Retrieval is preferred as the images annotated by humans manually by

entering keywords take more time. Content Based Image Retrieval employs different kinds of

query techniques such as Semantic Retrieval, Relevance feedback, Query by example,

Querying by image region, Querying by visual sketch etc. This thesis is focused on the

technique “Query by example” where the image retrieval system accepts an image as an input

and retrieves the similar images based on the comparison of features.

There are some challenging issues in content based image retrieval during the

similarity matching such as distance functions, semantic gap, and goal of users. Many

distance functions are available. Out of which the one that characterizes the underlying visual

similarity between the images has to be chosen. The gap between the image semantics and

the low-level features has to be bridged in image retrieval process. The goal of the user has to

be targeted effectively.



1.2 Existing System

Most of the approaches in image retrieval depend on the BOV model. BOV method

relies on quantifying the features extracted from images and indexing it using an inverted file

structure. But in some cases BOV method fails due to the error in quantization. These kinds

of problems will decrease the retrieval precision and recall due to the distinct features being

quantized to the similar visual word. This quantization may cause many false local matches

between images. It’s time complexity is also more.

1.3 Objective of the project

Now-a-days more and more images are available in web which can be retrieved by

efficient retrieval techniques. This thesis is focussed on searching and retrieving relevant

images from a huge database based on the interest points. It takes the spatial context

information along with the local features to increase the accuracy of image retrieval. Hessian

affine region detector is used to detect the interest points, corner, from the images. Scale

Invariant Feature Transform and Speeded up Robust Feature descriptors are used to extract

and describe the interest points from the images. The descriptors feature vectors extracted are

converted into binary codes. Geometric verification can be carried out by comparing the

binary codes. Indexing is done on the images in order to retrieve it from a huge database.

Indexing makes easier to find relevant images without having to search every image in the

database.

1.4 Proposed System

The proposed image retrieval system is capable of identifying similar images in the

database with high accuracy as it combines the spatial context information along with the

local features.

Hessian affine detector is used to detect the corner points in the images. Scale

Invariant Feature Transform (SIFT) and Speeded Up Robust Feature (SURF) descriptors are

used to detect and describe the interest points in the images. Combination of the descriptors

and detector will increase the retrieval rate.



The concept of proposed image retrieval system is as follows:

Figure 1.1: Proposed CBIR System

1.5 Applications of CBIR

Many possible applications for CBIR technology have been identified. Some of the areas are:

Crime prevention,

The military,

Fashion designing,

Journalism,

Medical diagnosis, and

Web searching.

1.6 Dissertation Organisation

The rest of this dissertation research is organised as follows: Chapter 2 summarises a

literature survey for Content Based Image Retrieval systems. Chapter 3 discusses the

requirement analysis, and specification. Chapter 4 focuses on the design of the image



retrieval system. Chapter 5 discusses the implementation details. Chapter 6 explains the

testing process that was carried out in the project. Chapter 7 focuses on the results of the

proposed system. Chapter 8 summarizes the project and discusses the work to be carried out

in the future.



CHAPTER 2

LITERATURE SURVEY ON IMAGE RETRIEVAL

Literature survey is one of the key steps in the project development, where good

amount of effort is put to learn and know the concepts, application, software and tools used,

drawbacks associated with each approach. After understanding the concept and latest trend,

the need for improvement is identified. Journals, web materials, etc. are used to learn and

understand the existing system.

2.1 Literature Review

Here different approaches [16] in image retrieval are discussed.

Figure 2.1 Approaches in CBIR system



Semantics-Sensitive Approach

In [6], the author explains an approach to Content-Based Image Retrieval which is

called Semantic-Sensitive approach [16]. In order to extract the feature correctly, a semantic

categorization is followed. Following this categorization, similarity is computed based on the

region. The important aspect of the proposed system is its retrieval speed. For similarity

matching purpose, the measure used here is integrated region matching which enables the

faster retrieval. This approach categorizes the images as either textured-non textured graph-

photograph. This approach was applied to a database with 200,000 images. Its retrieval of

images was accurate and fast. It was robust to intensity variations, sharpness variations,

scaling, rotation, and cropping. The limitation in this approach was that while classifying the

image, it may fall into second semantic classes.

Blobworld Representation of Images

In [2], [16], the author has explained how the querying is done on region using

homogenous colour texture segments known as blobs. It transforms the raw pixel data to a

smaller set of image regions that are coherent in colour and texture. This kind of image

representation is known as Blobworld representation which is created by clustering points in

a joint colour-texture position feature space. Image to Image matching is not followed. If the

blobs are identified by the user which is related to some concept such as “rose” then the users

search will be looking for a rose within other images in several different backgrounds. This

approach requires the involvement of the user. This algorithm has been run on 10,000 natural

images. This approach allows the user to view image’s internal representation and its results.

Content-Based Image Retrieval Using Shape and Depth features

In [1], [16], an algorithm for retrieving images using the shape information in an

image is discussed by the author. The 3D information of the image is also taken in to account.

This linear approximation procedure captures the depth information based on the idea of

shape from shading. The objects are retrieved using the similarity measure that combines

both the shape and the depth information. This approach has been effective in retrieving

engineering objects.



Spatial-Bag-of-Features

In [18], [16], the method of retrieving large scale images using a new bag-of-features

that includes the information regarding geometry of object is discussed by the author. The

local features of an image are projected to different directions or points in order to generate

the ordered bag-of-features. Based on this approach, different group of spatial-bag-of-features

are considered in order to capture the invariance of object’s geometry. A new bag-of-features

is generated by considering the selected representative features. This system leads to good

image matching and indexing for image retrieval and it also includes the spatial information

of local features to improve the retrieval accuracy. The author first evaluated the spatial bag-

of-features which was compared with bag-of-features, and then the bag of features was

compared with RANSAC re-ranking. This experiment was carried on Oxford5K dataset. To

test the effectiveness and scalability of spatial bag-of-features the dataset Panoramio1M was

leveraged.

Hashing Shape Context Descriptors

In [10], [16], a method for arranging and indexing the logo digital libraries is

discussed. The retrieval system compares the query image with the logo images present in the

database and retrieves them based on their similarity. These logos are described by a variant

of the shape context descriptor. The Locality Sensitive Hashing indexing structure is used to

arrange the descriptors to carry out the search process. Hashing techniques speeds up the

indexing and retrieves logos based on similarity. The author has carried out the experiment

on the tobacco-800 dataset to demonstrate the effectiveness and efficiency the proposed

method. The author has validated the approach by using a repeated random sub-sampling

validation scheme.

Bag of Visual Words

In [12], [16], the approach of Bag of Visual Words to retrieve the relevant word

images from a big database correctly is discussed. The approach discussed here uses the

principles of text retrieval system. The word images are represented in the form of histogram

of visual words. The histogram carries the information of the features in the image. Visual

words are quantized to represent local features in an image. Bag of Visual words method does

not explain the spatial relationship among visual words so the author has used re-ranking

method to the retrieved list of images in order to improve the performance. The author has



validated this approach on different Indian languages and has proved it, to be language

independent and scalable. The author has demonstrated the utility of the proposed system

across four Indian languages by using the dataset of 100K words. To demonstrate the

scalability, the author has used large dataset of 1M words. The performance is measures by

precision. The limitation of this process is the re-ranking step which is time consuming.

Wavelet Based Color Histogram Image Retrieval

In [9], [16], the author has explained the Content Based Image Retrieval using the

color and texture feature called Wavelet Based Color Histogram Image Retrieval. Wavelet

transformation is used to extract the texture features and color histogram is used to extract the

color features. The combination of both texture and color feature is robust to scaling and

translation of objects in an image. In order to extract the color features from digital images, a

clear picture about the representation of color in images is needed. The texture descriptor

provides measures for the properties such as smoothness, coarseness, and regularity. The

similarity between the images is computed using the distance functions. The author has

demonstrated the proposed retrieval method on a WANG image database with 1000 general

purpose color images and compared with the results of different authors. The result of the

proposed method was shown to have better performance than others with the average

retrieval time as 1 minute.

Principal Visual Word Discovery

In [17], [16], the author has described a method to detect license plates in various

observation angles, scale changes and illumination variation. It can also detect multiple plates

in the image. Scale Invariant Feature Transform descriptors are used to deal with different

angles, scale changes etc. The process is carried out by finding the visual word and matching

it. The author has demonstrated the proposed method on two different dataset such as LP

dataset and another dataset named Caltech Cars .The proposed approach is lower than the

other approaches in terms of detection rate. The other approaches are HLPE, LPE, and ESM.

A false positive rate of only 1.0% is only made by this method. Feature extraction time and

detection time are the two things that are considered by the author during investigating the

time efficiency. It takes less time in average to process one image. But the approach fails to

detect the license plate when the image quality is poor.



Comparison of SIFT and SURF

In [11], [16], SIFT descriptor is discussed which is used to describe the interest points

from the images , where features in the images remain the same whatever may be the image

scale and rotation and Speeded up Robust Features, a descriptor which is invariant to both

scale and rotation makes use of integral images. Image registration is a process that converts

the various data into a single coordinate system. It is a quite a complicated task in many

applications. In the process of Image registration, features are detected and matched as a first

step, then a transformation function is derived according to the features in the image and

finally the image is reconstructed based on the transformation function. For the experiment

purpose two images are taken by the author. Both the descriptors detect the features. The

author found out that SIFT descriptor detects more features when compared to SURF, but

SURF was found to be fast.

Scalable Partial-Duplicate Mobile Search

In [14], [16], the author has explained the large-scale partial duplicate image search

on the platforms of mobile. The SIFT descriptor, which is a histogram based descriptor is not

the best descriptor for the search so the author has proposed the Edge-SIFT (Scale Invariant

Feature Transform) descriptor. The descriptor is built with the help of edge maps by

considering both the location and orientation of edges. An inverted file based indexing

framework is said by the author in order to make use of Edge-SIFT descriptor. The Oxford

Building dataset is used in order to test the effects of different parameters and to evaluate the

validity of Edge-SIFT compression and compares with SIFT and ORB. Edge-SIFT perform

better than the other two approaches in terms of retrieval accuracy, efficiency, and

transmission cost and memory consumption. It cannot perform the tasks such as recognition

or classification.

Semantic-aware Co-indexing

In [13], [16], the author explains an algorithm called Semantic-aware Co-indexing

algorithm for vocabulary tree based image retrieval. It searches the images based on the

conditions such as the similarities between the images are based on the local features and

semantic attributes into the inverted indexes. After the search, the retrieval process considers

not only the images having same local features but also allows consensus in their semantic



similarities. The proposed algorithm changes the inverted indexes of local features which are

quantized by a large vocabulary tree. The methods used for the extraction of the BOW

features from the images are HOG and LBP, and the SIFT descriptors are extracted to use the

vocabulary tree. The online indexing of the features produced in this process is simple and it

is less memory consuming. The author has demonstrated the proposed approach on three

different datasets such as Holidays, UKbench, and Oxford. The overall discriminative

capability of the inverted indexes is increased by this system, which gives good retrieval

results.

Object based Image Retrieval using Combined Features

In [4], [16], local and global features are used to identify and extract the images. The

Bi-directional Empirical Mode Decomposition technique is used to detect the edges and

Harris Corner detector is used to detect the corner points of an image. HSV color feature is

used as the global feature. The author has applied the system on the ten categories of images

each with seventy two different orientations from COIL-100 image database.

HSV-Color Histogram and GLCM

In [3], [16], the Content Based Image Retrieval, which retrieves the images based on

the similarity of color and texture features of image sub blocks is discussed. The image is

segmented into sub-blocks of equal size. In order to extract the color from each sub block, the

HSV color space is quantified into non equal intervals and it is represented by cumulative

histogram. Gray-level occurrence matrix is used to get the texture of the sub-blocks. The

Similarity measure used here is Euclidean distance. This method has better performance than

the other system that uses only HSV color or GLCM texture or both HSV color and GLCM

texture.

Deep learning for Content-Based Image Retrieval

In [7], [16], the author has discussed a framework for CBIR .A large-scale deep

convolution neural networks is trained for learning feature representation of images .Deep

learning is an approach of machine learning. In this technique layers of information are

exploited for pattern classification for feature classification. The author has conducted

empirical studies for comprehensive evaluations of deep convolution neural networks with

applications to learn feature representations for different CBIR tasks under different settings.

The performance of feature representation scheme has been evaluated on new CBIR tasks



such as object retrieval using the Caltech256 dataset, facial image retrieval tasks using the

Pubfig83LFW dataset.

Summary

This chapter summarizes the state of the art methods and practices developed to

retrieve images. Most of the methods utilize image features like texture, color, shape, depth

for retrieving images. Methods which rely on color, texture fail to retrieve images taken at

different day light, views etc. Bag of visual word is an approach based on text retrieval

system, where vocabularies of image features are used for comparison and retrieval of

relevant images. Bag of Visual words method does not explain the spatial relationship among

visual words; hence accuracy of relevant image retrieval is low. SIFT descriptor based

retrieval system utilizes interest points and descriptors which defines the spatial relationship,

hence making it more efficient in retrieval. However SIFT is slow compared to other

methods. SURF is another approach based on feature vectors which is faster compared to

SIFT. Based on the literature study there is a need to develop new method for efficient image

retrieval system for the growing need of social media



CHAPTER 3

ARCHITECTURE OF CBIR SYSTEM

This chapter discusses the requirements and the overall design of the project. Design is an

explanation of the structure of the software that has to be implemented, and the algorithms

used. In the design phase, the translation of requirements takes place. The translation will be

a representation of software. The development of project is categorised into two activities

such as:

Top-level design: - It is also called as architectural design, which explains the

architecture of the system and identifies the various components. It also explains the

components relationship to one another.

Detailed design: - This explains each and every component in detail.

3.1 Requirement Analysis and Specification

The requirements are clearly understood and systematically organised in a specification

document called as Software Requirements Specification (SRS) document.SRS document

gives the overall behaviour of the system to be developed. The document includes the use

cases that describe how the user interacts with the software. The use cases are also called as

functional requirements. The document also maintains the non functional requirements which

impose the constraints on the design part or implementation. The constraints may be

performance requirements, quality standards or design constraints.

Functional Requirements

Functional Requirements defines how the system should respond to the given inputs or

conditions. The requirements may be calculations, processing or other functionalities that

defines what the system should accomplish. The following functional requirements are

identified in the system developed:

The system should provide a user interface for users to input the query image and

view the retrieved images.

The system should be able to upload the image from the dataset as per the user

request.



The system should be able to detect the interest points from the query image.

The system should be able to describe the detected interest points of the query image

by employing two descriptors.

The system should be able to combine the described interest points of the query

image.

The system should be able to convert the feature vectors into binary.

The system should be able to calculate the distance between the query image feature

and dataset image features for the similarity matching.

The system should be able to rank the images based on their similarity distance.

The system should be able to retrieve the images based on their ranking.

The system should provide an interface for the user to view the retrieved images.

Non Functional Requirements

The characteristics of the system are dealt by the non functional requirements. It could

be the constraints on the functions offered by the system. The characteristics are such as:

Reliability

It is the ability of a system to perform and maintain its required functions under

normal as well as unexpected conditions.

Robustness

It is the ability of the system to work well under ordinary as well as unexpected

conditions that stress its designer’s assumptions.

Usability

It is a qualitative attribute that measures how easy the user interface is to use. It is

associated with the functionalities of the system.

Portability

It is the software codebase feature which reuses the existing code instead of creating

new code when moving the software from an environment to another.

Performance

It is measured by the amount of work carried out by the system compared to the time

and resources used. It includes response time, completion time.

Cost

It is the value of money used to develop the software.



Scalability

It is the ability of the system to handle the growing amounts of work.

3.2 Architecture of the proposed CBIR system

Architectural design describes the high level overview of the decomposition of the

system. The decomposition is structural decomposition and functional decomposition. It

gives the information regarding the roles played by the system components. The Top level

design gives the System Block diagram which forms the blueprint of the system developed.

Figure 3.1 shows the high level architecture of the image retrieval system.

Development process involves the following steps:

User input

Pre Processing

Feature Extraction

Binarization

Similarity Computation and Ranking

Retrieval of images



Figure 3.1: Architecture of the proposed CBIR system

User Input

The user interface allows the user to input the query image and to visualise the similar

images that are being retrieved.

Pre Processing

Once the input is given, Pre-processing is done on the images at the lowest level of

abstraction. It aims at either modifying or enhancing the features of the image. Images are

processed, in order to represent the details of the image in numerical form.

Feature Extraction

The feature extraction subsystem extracts the interest points from the images, encodes

them into feature vectors. The encoded feature vectors are stored in the feature database,

which are further used for feature comparison. In this proposed system feature extraction



step is improved by incorporating SIFT, SURF descriptors and Hessian detector which

brings unique advantage to get improved image retrieval compared to existing methods.

Binarization

The query processing module extracts a feature vector from the query image, converts

them into binary as the binary codes are storage efficient and they require only few bits,

and makes the similarity process fast.

Similarity Computation and Ranking

The query processing module uses a distance measure (Hamming Distance) to find

the similarity between the query image and database images. Search results then can be

sorted based on the distance to the queried image.

Retrieval of Images

The images from the dataset relevant to the query image are retrieved based on the

similarity distance computed in the similarity computation and Ranking module.

3.3 Workflow of the system

This section gives the work flow of the CBIR system.

Figure 3.2: Work Flow of CBIR system



Summary

This chapter has covered the details of high level design such as system architecture

and workflow of the system. Using system architecture, step by step execution is explained

right from query image processing, database images processing, binarization, comparison and

retrieval. Feature extraction step is improved with the help of multiple descriptors which

makes the proposed system better on relevant image retrieval.



CHAPTER 4

LOW LEVEL DESIGN

The detailed low-level design provides the full details about the specification of the

design, for the system that will be developed.

4.1 Class Diagram

A class diagram describes the relationship between various classes and their

properties.

Figure 4.1: Class Diagram



4.2 Use Case Diagram

Use Case diagram describes the functionalities of the given system in terms of user,

target and interdependencies among the use cases. The use case diagram for the project is

shown in figure 4.2

Figure 4.2: Use-Case Diagram

The use cases for the proposed system are given above. Actor is the user who inputs the

query image, compares with the database images and retrieves the images relevant to the

query image.



4.3 Sequence Diagram

A Sequence diagram shows the operation of the processes with one another. It is said

to be an interaction diagram. It shows the order of the interaction and sequence of messages

between the processes.

Figure.4.3: Sequence Diagram

In the figure 4.3 , the vertical lines that are parallel indicate the different processes

involved in the proposed system and the horizontal lines gives the information that are

exchanged between those processes. The rectangles are called method call boxes that

represent the processes that are performed in response to the information.

Summary

This chapter gives an overview at low level design with the help of use case diagram,

sequence diagram and class diagram. Use case diagram visually explains the one to one

interaction between the user and modules such as the pre processing, query processing and

feature similarity module of the proposed method. With the help of sequence diagram, the

User Image Visual Content Feature Vector Similarity Result



order of the interaction and sequence of messages between the processes is explained. The

messages such as the features extracted from the images, the generated feature vectors, and

computed similarity distance is exchanged between the processes. Using class diagram,

functionalities of different classes such as the Hessian detector, SIFT and SURF descriptors

are explained.



CHAPTER 5

IMPLEMENTATION

The implementation phase gets the software or hardware operating system correctly in

its environment, including installation, configuration, running, and testing. This chapter

discusses the modules of the project such as pre processing, query processing, and similarity

and retrieval module.

5.1 Tool used for implementation

Appropriate tool has to be chosen while designing and building a software system in

order to implement the system. It will reduce the complexities in the coding and, testing part.

It will make the program readable one. It has good graphics capabilities. Matlab has become

such an important tool through the use of sets of its programs which supports a particular

work. These sets of programs are called toolboxes. The proposed system uses the image

processing toolbox.

In this project, MATLAB is used for implementing the design. Reasons for choosing it as

tool for implementation are listed below:

A high end numerical computing tool and environment for algorithm development,

for visualising and analysing the data

It manages codes, files, and data. It provides sophisticated platform to handle matrix

operations. Hence this is used in this project as a tool to handle digital images.

It can solve difficult mathematical problems for linear algebra, statistics, numerical

integration and differentiation

Custom GUI are build by the tools provided by MATLAB

Easy to integrate with other languages like C, C++, Fortran, Java, COM, and

Microsoft Excel.

5.2 Platform used for implementation

The application programs runs on the underlying computer system, the platform. In

this project, Windows 7 is used as a platform. Windows 7 is chosen as it is more flexible

when compared to Windows 95/98/XP. It has the capability of fast user switching that is, a



user can leave his desktop with all the applications open while the other user can use the

desktop with the same settings.

5.3 Module Implementation

This project has been implemented on Windows 7 platform using MATLAB. The

modules of the project are as follows:

5.3.1 Pre-Processing

Once the input is given, Pre-processing is done on the images at the lowest level of

abstraction. It aims at either modifying or enhancing the features of the image. Images are

processed, in order to represent the details of the image in numerical form.

5.3.2 Query Processing Module

The feature extraction subsystem extracts the interest points from the images, encodes

them into feature vectors and stores them into the feature database.

Corner refers to the position at which two lines, surfaces, or edges meet and form an

angle. Hessian affine region detector detects the interest points of the images which are

corners. First, the input image is smoothened with the help of guassian filter.The detector

finds the derivative points in the images. Feature points are obtained by comparing the

deriavative points with the pixels of the image.First, the images are initially divided into

small regions with particular window size. The interest points in the image are identified by

finding the derivatives of the particular region. The detector computes the second partial

derivative Ixx, Ixy, and Iyy for each image point. It also searches for the points where the

Hessian determinant is maximal.

The determinant of Hessian matrix [22] is given as below:

Ixx Ixy

H=

Ixy Iyy

Det (H) = IxxIyy – I^2xy -- (1)



The search is carried out by calculating the resultant image containing determinant

values. Later 3x3 search window is used to identify local maxima and in this process non

maximum points in the neighbourhood are suppressed. The search window is moved over the

entire image. The pixels with the larger value than the immediate neighbour values in that

window are alone kept. The detector gives all the remaining locations which are having a

value above a predefined threshold value. The resulting detector results are placed on corners.

Scale Invariant Feature Transform detects and describes the interest point of the

images and generates the descriptors which are feature vectors. First, the algorithm

smoothens the image using the Guassian filter.The first order and second order derivatives of

the image is computed. The Laplacian of Guassian[21] is calculated as

Δ2G = Ixx + Iyy -- (2)

The values are then arranged in a matrix format. The descriptor follows four steps to

extract the key points such as the detection of scale space extrema,localization of key points,

orientation assignment ,and key point descriptor generation.

Scale Space Extrema detection

Images are represented in various scales. The scale space representation is

parameterised by the size of the smoothening kernel.

Figure 5.1: Scale Space Extrema Detection [21]

The pixel X in the figure is compared with all the remaining 26 pixels in both the current

and adjacent scales. From this comparison, the pixel that is larger or smaller than all 26 pixels

will be choosen.



Keypoint localization

Extrema detection is done, from which, the points are taken. Since there are many

number of points, the locations of keypoints will not be accurate. The threshold value is

set in order to remove the outliers that is the weak points will be removed.

Orientation Assignment

Each keypoint is assigned an orientation. Relative to this orientation, the keypoint

descriptor is represented. The keypoint can also attain invariance to image rotation. The

derivatives, gradient magnitude, and direction of the smoothened input image is

computed at the scale of keypoint[21].

M(x,y) = sqrt ( Ix^2 + I y^2) -- (3)

θ(x,y) = tan-1 Ix/Iy --(4)

In a neighborhood of a key point, a wieghted direction histogram is created. Here

weights are represented as gradient magnitudes. The peak is selected as the direction of

the key point as depicted in the figure 5.2.

Figure 5.2: Orientation Assignment [21]

Keypoint Descriptor

At the keypoint, the relative orientation and magnitude in a 16 x 16 neighborhood is

calculated. For a 4 X 4 region, a weighted histogram is computed. The 16 histograms is

concatenated in one long vector of 128 dimensions. The numbers are hold on as feature

vector.



8 x 8 to 2 x 2 descriptors

Figure 5.3: Keypoint Descriptor [21]

In order to improve the point detection methods, the Speeded up Robust Feature

descriptor (SURF) is developed from the SIFT descriptors. The algorithm constructs the

integral image as the first step.

The integral image [33] is defined as follows:

I (x, y) = Input image x, y = spatial coordinate -- (5)

After the construction of the integral image, Hessian Matrix is obtained and Scale

Space is represented. The pixels are selected in order to generate the orientations.

Orientations are generated depending on the neighborhood of a particular interest point in

order to obtain a descriptor vector for every interest point. Thus the Key point descriptor is

generated.

The similarity between the images is found by combining and converting the obtained

feature vectors in to binary codes, since the binary codes will help in faster computation.

5.3.3 Feature Similarity Matching and Retrieval Module

The similarity between the query image and trained images is found by computing the

distance between them. The distance is computed by using a distance measure, Hamming

distance, in the proposed method. Hamming distance will find how far the binary codes of the

images are similar in terms of bit by bit. The results are sorted based on the minimum



Hamming distance. The retrieval module helps to select the images that will be presented to

the user as a result to query image based on the sorted distance.

Summary

This chapter has explained the steps involved in developing the proposed image

retrieval method. Major steps are interest point extraction and definition of feature vectors.

Hessian affine detector detects the corner points in the images. In SIFT method, interest

points are identified in scale space, key point localization, orientation assignment and key

point description. In SURF method, the integral image is constructed, interest points are

represented in scale space, hessian matrix is obtained and orientations are generated in order

to obtain the descriptor for every interest point. Calculated feature vectors of query image are

then used to compare with all feature vectors available in the data base. Image with minimum

hamming distance is identified as nearest match and the same image is retrieved. Entire

algorithm for this proposed method was developed using Matlab with user friendly Graphical

user interface.



CHAPTER 6

TESTING

Testing validates and verifies whether the software meets the requirements as stated

by the user. Testing helps us to find the correctness and completeness of the software. It also

ensures the quality of the software. There are two basic testing activities, such as component

testing and system testing. Component testing finds the defects by testing the individual

program components, whereas System testing tests whether the system meets the functional

and non functional requirements. A series of testing as discussed below is carried out for the

proposed system.

6.1 Unit Testing

Unit testing is a process of checking the individual modules to find whether they are

fit for use. A unit could be an individual function, procedure or method. It is also called as

module or component testing. It separates each part of the code and shows that they are

correct. Unit testing is carried out before integrating them into modules in order to check the

interfaces between the modules.

Table 6.1: Unit Testing Table

Name of the

Test Case

Test description Sample Input Expected

Output

Actual

result

Passed

(?)

Image

selection

A query image is

selected from a

set of images

User selects an

image

Selected Image

is displayed

As

Expected

Yes

Interest Point

Detection by

Detector

Interest Points

detection

Selected Image Detected

Interest Points

As

Expected

Yes

Feature vector

generation

Feature vector

generation

Interest points

detected from

the sample

image

Feature vectors

generated

As

Expected

Yes



Retrieval of

Images

Retrieval of

Images

Distance

computed for

the query and

dataset images

List of images

based on the

minimum

distance

As

Expected

Yes

6.2 Integration Testing

Integration testing is done, once when the unit testing of the modules is over. In

integration testing, the individual modules are tested first. Then they are tested as a group. It

takes those modules as input, that has been unit tested and aggregates them. This testing is

carried out to verify the functional and performance requirements.

Integration Testing Table

Table 6.2: Integration Testing Table

Modules Functions integrated Tests carried out

Query Processing Module Interest Point Detection

Feature Vector Generation

The module is tested to check

whether the interest points

are correctly detected and

described. It is also tested to

see whether the feature

vectors are correctly

generated.

Feature Similarity Matching

and Retrieval Module

Computation of distance of

query image

Retrieval of images based on

minimum distance

The module is tested to check

whether the retrieved list of

images is relevant to the

query image.

6.3 System Testing

System testing is carried out on the integrated system in order to check the

compliance of the system with its requirements. The components that were successful in

integration testing are the inputs for system testing. The purpose of system testing is to check

whether there are any inconsistencies between the software units that were integrated

together. It also validates whether the system meets the specified requirements. It also checks

the design, behaviour and the expectations of the user.



System Testing Table

Table 6.3: System Testing Table

Functionality Test Input Tests conducted

Working of Proposed System

Image

Functional requirement test

Summary

The proposed system has been tested rigorously at different levels right from unit

testing up to system testing. At unit level key tests like query image selection, corner

detection, interest point detection, image display, and similarity distance calculation, relevant

images retrieval were checked and ensured. Subsequent to unit level testing, integration tests

were also carried out to eliminate errors due to integration of modules. All these tests and

checks confirmed the compliance of functional requirements of the proposed module



CHAPTER 7

RESULTS AND DISCUSSION

The tool MATLAB R2013a (8.1.0.604) is used to carry out the experiment. For this

experiment, a dataset with one hundred and fifty images in JPEG format of size 256 x 256 is

employed as shown in figure 7.1. Searching of images relies on the similarity means rather

than the actual matching. Every query image returns the top 10 closest matching images.

The dataset used for the experiment contains 150 images which are shown below:



Figure 7.1 Dataset used for the proposed system

The following are the different categories of images in the dataset:



Figure 7.2 Different categories of images used for the proposed system

The evaluation measures for the proposed system are Recall and Precision [23].

Precision [23] is the ratio of relevant images to total retrieved images as given in equation

7.1, whereas Recall [23] is the ratio of appropriate images to total relevant images as given in

the equation 7.2. Here appropriate images refer to the relevant images. The mean average

precision and recall values for the proposed system are given in table 7.2.

Precision: A -- (6)

C

A: Number of relevant images retrieved

C: Total Number of images retrieved

Recall: A -- (7)

B

A: Number of relevant images retrieved

B: Total Number of relevant images

Mean Average Precision: ∑ Precision -- (8)

In the proposed system, the distance between the feature vectors of given image and

database is computed in order to find the similarity between them based on the minimum

distance. The distance measure used in this work is Hamming distance.

Hamming distance = √ ((mean (feature vectors of query image)-mean (feature vectors of dataset image)) ^2

-- (8)



The Precision and Recall values for different categories of images are:

Table 7.1: Precion and Recall for each category

Category Precision Recall

Taj Mahal 0.94 0.63

Ice Berg 0.58 0.56

Pyramid 0.78 0.52

Colosseum 0.77 0.51

Eiffel Tower 0.73 0.48

Christ the Redeemer 0.70 0.47

Harmandir Sahib 0.95 0.63

Mahablipuram 0.60 0.40

Victoria Palace 0.67 0.44

Iskcon Temple 0.74 0.49

The graph showing the precision and recall values for different categories of images are:

Graph 7.1: Precision and Recall for each category

The precision level for each category is in the order of 0.7 and recall in the order of 0.5.The

proposed method is capable of producing high level of precision and recall even for wider



range of images starting from taj mahal to iskcon temple with different views , colors and

depth.

The below image indicates the retrieved similar images for the given input image:

Figure 7.3: Retrieved list of images



The Precision and the Recall measure for the proposed method is as follows:

Table 7.2: Mean Average Precion and Recall

Method Mean Average Precision Recall

Existing method 0.67 0.45

Proposed Method –

SIFT+SURF+Hessian

0.74 0.51

Comparison Graph

The proposed algorithm was tested for 150 images. The Mean Average Precision was

found to be increasing by 7% and Recall was found to be increasing by 6% when compared

to the existing method because of two descriptors and one detector being employed in this

work in order to detect, describe the interest points and spatial context information.

Graph 7.2: Mean Average Precision and Recall

Summary

The proposed method retrieved relevant images for the given query image accurately.

The performance has been evaluated using several test images ranging from Taj Mahal , Ice

Berg, Pyramid, Colosseum, Eiffel Tower, Christ the Redeemer, Harmandir Sahib,

Mahablipuram, Victoria Palace and Iskcon Temple with different views, colours and depth.

Mean average precision of the proposed method is 0.74; recall is 0.51, which are higher than



the existing method [19] by 7%. This proves the proposed approach of SIFT+SURF+Hessian

has better retrieval accuracy compared to other techniques even with significantly different

angles and daylight colours.



CHAPTER 8

CONCLUSION AND FUTURE WORK

CONCLUSION

Accurate image retrieval is one of the growing needs in order to make use of the huge

number of images that are available in social/web media. There are several methods

developed for image retrieval right from techniques like text based, local and global features

like colour, texture, shape etc. Every method has its own merits and de-merits. In this project

a new approach is proposed which is based on SIFT, SURF descriptors and Hessian detector.

This approach is based on identification of interest point with spatial context information and

generation of feature vector. These feature vectors of query image and data base images are

used for comparison and retrieval with the help of minimum hamming distance. Since the

proposed method does not rely on color or texture, this can accurately retrieve similar images

with different angle and daylight exposure. This proposed method was developed and tested

using Matlab and accuracy of the system was experimented with 150 significantly different

set of test images. Results were found better than existing method [19] by more than 7% in

terms of mean average precision and recall.

FUTURE WORK

In future, work can be carried out in reducing the number of features in images, with

the same retrieval performance. Also research can be done on image interpretation, text

retrieval from images and interpretation as a next step of similarity matches. The experiment

need to be tested in a very large database in real-time web application.



REFERENCES

[1] Amit Jain, Ramanathan Muthuganapathy, and Karthik Ramani, “Content-Based

Image Retrieval Using Shape and Depth from an Engineering Database”, G. Bebis et

al. (Eds.): ISVC 2007, Part II, LNCS 4842, pp. 255–264, 2007.

[2] C. Carson, S. Belongie, H. Greenspan, and J. Malik, “Blobworld: Image

Segmentation Using Expectation-maximization and Its Application to Image

Querying”, IEEE Trans. Pattern Analysis and Machine Intelligence, 24(8):1026-1038,

2002.

[3] Deepak John, S.T.Tharani, and SreeKumar.K, “Content Based Image Retrieval using

HSV-Color Histogram and GLCM”, An International Journal (IJARCSMS), Vol.2,

Issue No.1, January 2014.

[4] H.Kavitha, M.V.Sudhamani, “Object based Image Retrieval from Database using

Combined Features”, International Journal of Computer Applications, Vol.76, Issue

No.8, August 2013.

[5] J. Sivic and A. Zisserman, “Video google: A text retrieval approach to object

matching in videos”, in Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR),

Dec.2003, pp. 1470–1477.

[6] J.Z. Wang, J. Li, and G. Wiederhold, “SIMPLIcity: Semantics-Sensitive Integrated

Matching for Picture Libraries”, IEEE Trans. Pattern Analysis and Machine

Intelligence, 23(9), 947–963, 2001.

[7] Ji Wan, Dayong Wang, Steven C.H. Hoi, Pengcheng Wu,Jianke Zhu, Yongdong

Zhang, Jintao Li,” Deep Learning for Content-Based Image Retrieval: A

Comprehensive Study”, in Proc. ACM Int. Conf. Multimedia, 2014.

[8] Lowe, D. “Distinctive image features from scale-invariant keypoints” International

Journal of Computer Vision, 60, 2 (2004), pp. 91-110.

[9] Manimala Singha and K.Hemachandran, “Content Based Image Retrieval using Color

and Texture”, Signal & Image Processing: An International Journal (SIPIJ) Vol.3,

No.1, February 2012.

[10] Marçal Rusinol, Josep Llados,“ Efficient Logo Retrieval through Hashing Shape

Context Descriptors”, in Proc. ACM, 2010.

[11] P M Panchal, S R Panchal, S K Shah,” A Comparison of SIFT and SURF”, IJIRCCE,

Vol.1.Issue 2, April 2003.



[12] Ravi Shekhar and C.V. Jawahar, “Word Image Retrieval using Bag of Visual Words”,

in Proc.IEEE, 2012.

[13] S. Zhang, M. Yang, X. Wang, Y. Lin, and Q. Tian, “Semantic-aware co-indexing for

image retrieval”, in Proc. IEEE Int. Conf. Comput. Vis. (ICCV), Apr. 2013, pp. 1–8.

[14] S. Zhang, Q. Tian, K. Lu, Q. Huang, and W. Gao, “Edge-sift: Discriminative binary

descriptor for scalable partial-duplicate mobile search,” IEEE Trans. Image Process.,

vol. 22, no. 7, pp. 2889–2902, Jul. 2013.

[15] Subathra Muthuraman, Mrs.Swathi Baswaraju, Mrs.B.Mounica, “Large Scale Image

Retrieval using Descriptors and Distance Measure”, IJCSMC, Vol.4, Issue No.5, May

2015.

[16] Subathra Muthuraman, Mrs.Swathi Baswaraju, Mrs.B.Mounica, “Survey on

Contextual Hashing for Searching and Retrieval of Images using Descriptors and

Distance Measure”,IJERT,Vol.4,Issue No.1,Jan 2015.

[17] W. Zhou, H. Li, Y. Lu, and Q. Tian, “Principal visual word discovery for automatic

license plate detection”, IEEE Trans. Image Process., vol. 21, no. 9, pp. 4269–4279,

Sep. 2012.

[18] Y. Cao, C. Wang, Z. Li, L. Zhang, and L. Zhang, “Spatial-bag-of features”, in Proc.

IEEE Conf. Comput. Vis. Pattern Recognit., Jun. 2010, pp. 3352–3359.

[19] Zhen Liu, Houqiang Li, Wengang Zhou, Ruizhen Zhao, and Qi Tian,“Contextual

Hashing for Large-Scale Image Search”, IEEE transactions,VOL.23, No.4, April

2014.”

[20] http:// en.wikipedia.org/wiki/Content-based_image_retrieval [ July 2014]

[21] http://crcv.ucf.edu/people/faculty/shah.php [ August 2014]

[22] http://en.wikipedia.org/wiki/Hessian_affine_region_detector [August 2014]

[23] http://en.wikipedia.org/wiki/Precision_and_recall [ February 2015]

[24] http://en.wikipedia.org/wiki/Scale_space [ August 2014]

[25] http://ijarcet.org/wp-content/uploads/VOLUME-2-ISSUE-6-2077-2080.pdf

[26] http://in.mathworks.com/products/matlab [ September 2014]

[27] http://in.mathworks.com/support/install-matlab.html [September 2014]

[28] http://www.academia.edu/5089977/A.Net_Implementation_of_a_Content-

Based_Image_Search_Component [ December 2014]

[29] http://www.ijrat.org/downloads/may-2014/paper%20id-25201471.pdf [December

2014]



[30] http://www.researchgate.net/publication/264276256Image_Retrieval_A_Literature_

Review [ December 2014]

[31] http://www.scribd.com/doc/28209229/CBIR-Content-Based-Image-Retrieval#scribd

[ March 2015]

[32] http://www.uni-weimar.de/medien/webis/publications/papers/stein_1999a.pdf

[December 2014]

[33] www.vision.ee.ethz.ch/~surf/eccv06.pdf [July 2014]



Appendix-A

ACRONYMS

The acronyms used in the project are:

CBIR - Content Based Image Retrieval

SIFT -Scale Invariant Feature Transform

SURF -Speeded Up Robust Feature

BOV - Bag of Visual words

RANSAC - Random Sample Consensus

LOG - Laplacian of Gaussian

SRS -Software Requirement Specification

QBIC - Query by Image Content

HOG - Histogram of Oriented Gradient

LBP - Local Binary Pattern

MATLAB -Matrix Laboratory

BOW -Bag of Words

GUI -Graphical User Interface

GLCM - Gray Level Co-occurrence Matrix

Readme

The experiment is carried out with the following requirements:

Processors : Intel Pentium

RAM : 2 GB.

Storage : 1GB.

Platform : Windows 7

Software : MATLAB R2013a (8.1.0.604)

Installation Procedure for MATLAB software [32]

The following files have to be downloaded [32]:

win32.rar or win64.rar (for 64 bit system)

PLP



"license.dat" into a folder.

The steps [32] for installation are as follows:

Step 1: First, the setup.exe file has to be double clicked

Step 2: Once, the setup file is clicked, the installer shows the dialog box, from which the

option install without using the internet [32] has to be chosen.

Step 3: The license agreement has to be accepted.

Step 4: The file installation key has to be entered.

Step 5: The installation type has to chosen, either the typical or custom type.

Step 6: The folder has to be specified, where the Matlab has to be installed.

Step 7: The location of the license file has to be specified.

Step 8: After specifying the location of the license file, press Next to continue the installation.

Step 9: The files has to be copied to the drives.

Step 10: The installation is complete.

Step 11: Add the environment variables.



Appendix-B

SNAPSHOTS

The following snapshots define the results or outputs that we will get after step by

step execution of all the modules of the system.

Figure A.1: Main Screen



When the button named input image is clicked, the below screen appears to select the

query image.

Figure A.2: Dataset Images



After selecting the input image , the interest points are detected as below:

Figure A.3: Detected Interest Points



When the button named combined features is chosen, the detected interest point by

the detector and the feature vector of the interest point generated by the descriptors are

combined together and displayed as follows:

Figure A.4: Combined Features



The detected interest points by the Hessian affine detector are as shown as below:

Figure A.5: Hessian Detector Points



The detected interest points by the SIFT descriptor is shown as below:

Figure A.6: SIFT Descriptor Points



The SURF descriptor points are shown as below:

Figure A.7: SURF Descriptor points



The spatial context information of the interest points are shown as below:

Figure A.8: Spatial context information – Hessian Points

Figure A.9: Spatial context information – SIFT Points



Figure A.10: Spatial context information – SURF Points

When the button named Retrieval is chosen, the list of images relevant to input image

appears with the similarity distance.

Figure A.11: Result Display



Appendix-C

PAPER PUBLISHED IN INTERNATIONAL JOURNAL OF

ENGINEERING RESEARCH & TECHNOLOGY



PAPER PUBLISHED IN INTERNATIONAL JOURNAL OF

COMPUTER SCIENCE AND MOBILE COMPUTING

VISVESVARAYA TECHNOLOGICAL UNIVERSITY BELAGAVI ...

Documents

Transcript of VISVESVARAYA TECHNOLOGICAL UNIVERSITY BELAGAVI ...