VISVESVARAYA TECHNOLOGICAL UNIVERSITY

“JnanaSangama”, Belagavi, Karnataka-590018

2022 - 2023
A Project
Report On
“Tobacco Plant Region Extraction and segmentation using CNN
and watershed algorithm”
Submitted in partial fulfillment of the requirements for the award of the degree of
Bachelor of
Engineering in
Computer Science and Engineering
Submitted by

Gayathri M 4VM19CS018
Abhishek S 4VM20CS400
Harsha M 4VM20CS401
J Sumanth 4VM20CS403

Under the Guidance of

Internal guide External guide

Prof Rumana Anjum Shridhar Patil

Assistant Professor Developer
Dept. of CSE, VVIET Mindset solutions

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

VIDYA VIKAS INSTITUTE OF ENGINEERING & TECHNOLOGY
#127-128, Mysore - Bannur Road, Alanahally, Mysuru, Karnataka 570028
 idya Vikas Educational Trust ®

VIDYA VIKAS INSTITUTE OF ENGINEERING &

TECHNOLOGY
#127-128, Mysore - Bannur Road, Alanahally, Mysuru, Karnataka 570028

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

CERTIFICATE

Certified that the 8th Semester Project titled “Tobacco Plant Region Extraction and segmentation
using CNN and watershed algorithm” is a bonafide work carried out by Gayathri M
(4VM19CS018), Abhishek S (4VM20CS400), Harsha M(4VM20CS401) and J
Sumanth(4VM20CS403) in partial fulfilment for the award of degree of Bachelor of Engineering in
Computer Science and Engineering of the Visvesvaraya Technological University, Belagavi, during the
year 2022-23. The Project report has been approved as it satisfies the academic requirements with respect to
the project work prescribed for Bachelor of Engineering Degree.

Signature of the Guide Signature of the HOD Signature of the Principal

Prof. Rumana Anjum Dr. Madhu B K Dr. Manjunatha T S
Assistant Professor Professor & Principal, VVIET
Dept. of CSE Head Dept. of Mysuru
VVIET, Mysuru CSE
VVIET, Mysuru

External Viva

Name of the Examiners Signature with Date

1.

2.
 ACKNOWLEDGEMENT

The joy and satisfaction that accompany the successful completion of any task would be incomplete
without the mention of the people who made it possible.

First and foremost we offer our sincere phrases of thanks to Sri. Vasu, Chairman of VVET,
Mr. Kaveesh Gowda V, Secretary of VVET and all management members of VVIET, Mysuru for
providing help and support to carry out the internship.

We would like to express our gratitude to our beloved Principal, Dr. Manjunatha T S for
providing us a congenial environment for engineering studies and also for having showed us the
way to carry out the internship.

We consider it a privilege and honour to express our sincere thanks to Dr. Madhu B K
Professor and Head, Department of Computer Science and Engineering for his support and
invaluable guidance throughout the tenure of this internship.

We would like to thank our Guide Prof. Rumana Anjum, Assistant Professor, Department
of Computer Science and Engineering for her/his support, guidance, motivation, encouragement for
the successful completion of this internship.

We intend to thank all the teaching and non teaching staffs of our Department of Computer
Science and Engineering for their immense help and co-operation.

Finally we would like to express our gratitude to our parents and friends who always stood
by us.

Regards
Gayathri M(4VM19CS018)

Abhishek

S(4VM20CS400) Harsha

M(4VM20CS401) J

Sumanth (4VM20CS403)
 ABSTRACT

Introduction

When an item or person is shown in the form of an image for example, via photograph or by
two-dimensional painting—the object or person's look is sent to the viewer, offering a visual
representation of the thing or person. Picture processing considers an image to be an
amplitude distribution of color (s). Photos, screens, holograms, and other three-dimensional
displays are all examples of two-dimensional images. Figure shows an example of a picture.
Any two-dimensional figure, such as a map, graph, pie chart, or painting, might be
considered a "image." Images may be created in a variety of ways, including by hand, such as
by drawing, painting, or carving, as well as mechanically, using printing or computer
graphics technology, or by combining these techniques, as in a faux photograph. For a brief
period, a volatile picture is one that does not last long. A mirror reflection, a camera obscura
projection, or a cathode ray tube picture are all possibilities. As the name implies, a fixed
image is a digital or photographic picture that has been recorded on a physical medium such
as paper or cloth. Using computer algorithms to handle digital photographs is known as
digital image processing in computer science. As a subset of digital signal processing, digital
image processing has significant advantages over analog image processing. Because there are
so many different algorithms that may be used to handle input data, problems like noise
build-up and signal distortion are far less likely to arise during processing. Pictures may be
characterized as multidimensional systems since they are defined in at least two dimensions.

Project Methodology

In order to produce a product or solve an issue, the system development technique must be
followed. There are many stages, techniques, and actions involved in creating a piece of
software. For product development, it follows a set of stages. The waterfall model is used in
this project's development.
Requirement Analysis: This phase focuses on gathering the system's requirements.
Documentation and requirements review are part of this step.

System Design: The system specs are converted into a software representation with the
needs in mind. Algorithms, data structures, and software architecture are some of the topics
covered in this phase.
Coding: In this phase, the programmer begins coding in order to sketch out the product's
features in complete detail. Machine-readable code is everything that is generated from system
requirements.

Implementation: Software coding and programming are both part of the implementation
process. Typically, the product of this phase includes the library, executables, user guides, and
other documentation for the program.

Testing: In this step, each program (model) is put together and tested to verify that the
software requirements are met. Verification and validation are the primary goals of testing.

Maintenance: Customers' needs vary over time; external environments change; faults and
oversights discovered during testing are fixed; and software efficiency is improved throughout
the maintenance phase of development.

Conclusion and Application

Effective management of plant species in agriculture requires reliable plant

recognition capabilities, while botanists often use such applications for studying
medicinal purposes. In our project, classification refers to assigning a specific plant
species to an image by analyzing extracted features. Essentially, classification is
accomplished by utilizing prior knowledge to identify class labels associated with
new input images. Our system aims at accurately recognizing various types of
plants in agricultural settings with an emphasis on speed and accuracy during
detection processes. Future efforts will concentrate on developing advanced
algorithms that deliver quick and precise results during the detection phase.
 INDEX
SL NO CHAPTERS PAGE NO

INTRODUCTION
1 1.1 OVERVIEW 1-3
1.2 OBJECTIVE 3
1.3 EXISTING SYSTEM 3
1.4 PROPOSED SYSTEM 4-5
LITERATURE SURVEY
2 2.1 LITERATURE SURVEY 6-7
L.MENG AND J.P.KEREKES

SOFTWARE REQUIREMENT SPECIFICATION

3
3.1 FUNCTIONAL REQURIMENT 8
3.2 NON-FUNCTIONAL REQURIMENT 9-11
3.3 RESOURCE REQUIREMENT 11-12
3.4 HARDWARE REQUIREMENTS 13
SYSTEM ANALYSIS
4 4.1 FEASIBILITY STUDY 14-15
SYSTEM DESIGN
5.1 FUNDAMENTAL DESIGN CONCEPT 16-20
5.2 SYSTEM DEVELOPMENT METHODOLOGY 20-22
5 5.3 SYSTEM ARCHITECTURE 22-23
5.4 USE CASE DIAGRAM 25
5.5 SEQUENCE DIAGRAM 25-26
27-28
5.6 DATA FLOW DIAGRAM
IMPLEMENTATION
6 6.1 LANGUAGE USED 29-30
6.2 METHODOLOGY 31
6.3 CODE SNIPPETS 32-33
6.4 IMPLEMENTATION 34-41
7 SOFTWARE TESTING
7.1 UNIT TESTING 42-44
7.2 TESTING AND VALIDATIONS 44-49
8 CONCLUSION 50
9 BIBILOGRAPHY 51
Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Chapter 1

INTRODUCTION
When we utilize photographs or 2D paintings for representation purposes with regards to objects or
people, we offer viewers with the chance to perceive their appearance through visual depictions in
the form of images. Amplitude distribution is used in pictures processing as one way for determining
their coloring effect.2D imagery is exemplified by various presentations such as photos, screens and
even holograms - all created with three-dimensions in mind. The figure displayed below highlights
one clear case-in-point.

An image encompasses any two-dimensional form - from maps to graphs to pie charts and paintings.
While some are crafted by hand through techniques such as drawing or painting; others are
mechanized with the employment of printing or computer graphics technologies. In some instances,
however, these methods are combined for example when creating faux-photographs. The visual
realm can be broken down into two fundamental types: volatile and fixed images. Volatile images
are those that disappear quickly after coming into existence – examples could be mirror reflections or
cathode ray tube pictures. Conversely fixed images have been saved onto tangible mediums such as
paper or cloth. Digital image processing involves using computer algorithms to manage photographic
content – its a subset of digital signal processing within computer science and offers distinct benefits
over analog methods due to its diverse range of algorithms available for managing input data. With
these multiple algorithms issues like noise buildup and signal distortion and far less likely to occur.

Dept of CSE, VVIET 2022-23 Page | 1

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Imaging Devices

The creation of images through various imaging gadgets is an essential part of modern scientific
research. The pioneers like Galileo Galilei and Newton aimed at designing & enhancing image
technologies in their time. Today there are multiple imaging devices like

 eye-glasses
 camera obscura,
 magnifying glasses;
 microscope;
 telescope;
 holography available.

Features of Digital Imaging & Image Processing

Digital imaging systems have advanced significantly owing to the presence of computers &
processors today making them highly versatile. One key advantage offered by digital computers over
traditional analog electrical & optical information processing systems is the ability to reprogram
without any hardware changes for undertaking new tasks. Simply developing a suitable computer
code can enable you to create complex problem solvers using the same resources- hardware or
software that were used earlier too. The swift adaptation capability towards different signal types as
per user requirements makes computers an effective tool for adaptive image processing. Signal
convolution and correlation in analog optics involve transformations such as spatial and temporal
Fourier analysis. Nonetheless, imaging systems that incorporate computers can perform any
operation. Optoelectronic information processing is now extremely powerful when combined with
digital signal processing.

Objectives:

The ability to identify and recognize specific objects or groups within images serves as an integral
component across numerous applications.

Dept of CSE, VVIET 2022-23 Page | 2

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Existing System:

The necessity for identifying objects or groups within photographs reaches varying industries such
as agriculture and surveillance sectors. Developments have been made to automatically recognize
photos containing members of an object class via research findings. It is worth noting that
appropriate attention is critical when identifying the tobacco plant.

Proposed System:

In this study, we devised a three-phase system commencing with morphological procedures followed
by watershed segmentation for potential tobacco plant regions via UAV photos. The creation and
training of a deep convolutional neural network (CNN) serve as the subsequent step responsible for
distinguishing between distinct types of tobacco plants while removing non-tobacco plants through
post-processing measures.

Tobacco plant area extraction comprises four steps including

 noise filtering;
 soil region detection;
 plant region segmentation;
 and finally, plant region extraction.

Dept of CSE, VVIET 2022-23 Page | 3

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Fig1.2: Framework for Plant region extraction

Tobacco plants may be detected in UAV photos using a novel technique based on deep neural networks.
According to our knowledge, this is the first study to look for tobacco plants in photographs captured by an
unmanned aerial vehicle (UAV). A dataset of UAV images is used to test the effectiveness of the
algorithm under consideration.

Dept of CSE, VVIET 2022-23 Page | 4

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Chapter 2

LITERATURE SURVEY

L. Meng and J. P. Kerekes, “Object tracking using high resolution satellite imagery,”[1]

There are several uses for high-resolution multispectral satellite photos that may be seen from multiple angles.
A three-step system for monitoring moving objects, modeling targets, and finding matches for those targets is
presented. The time-series photos are used to identify items that may be moving. Next, spectral, and spatial
information are extracted from the target to create a model. As part of the matching process, a new regional
operator is created by combining the Bhattacharyya distance with the histogram intersection and pixel count
similarity. The WorldView-2 satellite provided us with a series of multi-angle sequence photos, which we used
to test our technique. The test's recall, precision, and F1 score are used to evaluate the tracking performance.
Using high-resolution multispectral satellite images, we were able to show object tracking in a challenging
setting in this research.
JIANG HUIXIAN, The Analysis of Plants Image Recognition Based on Deep
Learning and Artificial Neural Network [2]
In order to better understand and protect the plants that we come into contact with on a daily basis,
categorization and identification of plants is essential. When identifying a plant's identification, it is important
to look at its leaves. Artificial intelligence and machine vision are being used to improve plant classification
and protection using image analysis-based plant leaf identification technologies. "Deep learning" is the
shorthand for neural networks at the heart of deep learning. Plant leaf samples are automatically trained and
categorized using an artificial neural network that uses the back propagation approach. Image analysis will be
used to extract leaf characteristics and identify plant species as the major objective of this project. Images of
plant leaves are first segmented using a variety of techniques before being subjected to an algorithm for
extracting the form and texture of individual leaves. Data on the chemical characteristics of plant leaves is then
collected and organized in line with get all the relevant information that is out there. KNN-based neighborhood
classification, SVM-based support vector machine, and Kohonen network based on self-organizing feature
mapping are used to investigate and compare 50 plant leaf datasets. In addition, ginkgo leaves were shown to
be simpler to recognize from those of seven other kinds of trees. Leaf photographs with a complex background
have an excellent recognition effect. Using photos from the test set, the learning model calculates
reconstruction errors. Recreating a deep learning model with the smallest number of errors will provide the test

Dept of CSE, VVIET 2022-23 Page | 5

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm
set's class label. In terms of recognition time, we

Dept of CSE, VVIET 2022-23 Page | 6

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

discovered that this strategy is the quickest and most accurate.

Summary
This chapter focuses on the publications and resources that were consulted while writing this dissertation. All
of these articles and websites discuss how to teach group conduct, as well as the many approaches that have
been tried, as well as the benefits and drawbacks of each.

Dept of CSE, VVIET 2022-23 Page | 7

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Chapter 3

SYSTEM REQUIREMENT SPECIFICATION

The foundation of an effective software development process lies in its specifications that outline all critical
elements and needs of the system. Software requirements specification (SRS), in particular, highlights a
company's preliminary comprehension of their customer or prospect's demands and dependencies on paper.
Both parties need to have an open line of communication for smooth collaboration throughout the project
lifecycle while using SRS as a cost control mechanism. Besides acting as a primary source document for
developers, SRS is also known as its parent document because it links other documents such as design specs,
software architecture specs, testing plans, validation plans documentation plans and statements of work with
ease. However, one must know that an SRS does not cover anything beyond what developers perceive their
clients' system needs.

3.1 Functional Requirement

When a system is provided with specified inputs or situations, a Functional Requirement specifies how the
system should respond. Some examples of these are computations and data manipulation and processing. In this
system, the functional requirements are as follows:-

1. Load the dataset into memory.

2. Load the picture from the input.
3. For segmentation, use Watershed.
4. Using CNN, train a dataset.
5. Using CNN, classify the picture.

3.2 Non-functional Requirement

Requirements that do not directly affect the system's operation are known as non-functional requirements
(NFR). A system's performance is judged by its ability to meet a set of predetermined criteria, rather than by its
unique actions. It's possible that they have anything to do with emergent system characteristics like
dependability, reaction speed, and store occupancy. Non-functional requirements may be triggered by a variety
of circumstances, including those related to the demands of the end user, budgetary restrictions, corporate

Dept of CSE, VVIET 2022-23 Page | 8

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

regulations, and the desire for compatibility with other software and hardware systems.
1. Requirements for the product
2. Requirements of the Organization
3. A user's needs
4. Requirements for a basic operation
We have established several requirements to ensure our product meets user expectations:
Portability – Our program is built using Python ensuring its compatibility on any platform supporting said
language.
Correctness – We pride ourselves in our meticulous approach utilizing well-defined criteria and methods
resulting in accurate data output after rigorous testing.
Ease of Use – Our intuitive front-end interface facilitates seamless interaction without compromising
functionality or productivity.
Modularity – Breaking down our software into smaller modules enhances flexibility while maintaining smooth
interfacing via defined interfaces.
Robustness – Our aim is achieving peak performance while still optimizing speed without sacrificing accuracy.
Python's robust attributes lower the system failure rate which belongs under non-functional needs further
classified into execution quality or evolution quality subcategories.
Aspects like testing, maintenance, scalability, or expandability are part of evolution quality which can be
observed during development. University educated individuals possess a higher level of cognitive ability and
analytical skills making them more capable of comprehending complex ideas and engaging in critical thinking.
This is particularly important in today’s society, where issues are becoming increasingly complex and
multifaceted. As such having a university education is beneficial for both personal development and career
success. Moreover research has consistently shown that university graduates earn higher salaries than those
without a degree. According to the US Census Bureau individuals with a bachelors degree earn on average
$30,000 more per year than those with only a high school diploma. This disparity only widens as one climbs
the educational ladder.
However it is not just about financial gain. University education also provides individuals with opportunities
for personal growth and self discovery. Exposure to diverse perspectives and experiences can broaden ones
worldview and lead to greater empathy and understanding towards others. In summary pursuing higher
education has numerous benefits that extend beyond career advancement. It fosters critical thinking skills
increases earning potential and promotes personal growth.

Dept of CSE, VVIET 2022-23 Page | 9

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

If you're into scientific programming with Python Anaconda has got you covered. It offers Spyder, an IDE
designed specifically for this purpose. Not only that, but it also comes with an introspective and interactive
testing environment thats part of the package. Once you've installed Anaconda simply run the command spyder
to launch Spyder on Windows Mac OS X, or Linux. Alternatively, if you prefer using Anaconda Navigator
(also included with Anaconda) you can find Spyder on the Navigation Home tab there. To learn more about
Spyder and its capabilities visit either the official Spyder website or consult the Spyder documentation. And
speaking of command prompts: did you know that just like the regular command prompt Anacondas own
version lets you run anaconda and conda commands without having to change folders or modify your path?
This makes executing commands and scripts a breeze.
MATLAB® on the other hand is a programming language that directly expresses matrix and array
mathematics within its desktop environment. This makes it ideal for iterative analysis or design processes.
Seamlessly integrate code, output, and formatted text into one notebook with the help of the Live Editor.

Our MATLAB toolboxes are expertly crafted, meticulously tested, and extensively documented to guarantee
superior quality. Discover various algorithms through our interactive apps that enable you to explore data
interactions. Boost productivity by automating or repeating tasks using MATLAB's programming capabilities
upon achieving optimal outcomes.

Python Programming

As far as general-purpose high-level programming languages go few surpass Pythons popularity and
effectiveness. Designed by Guido van Rossum in 1991 with the goal of achieving superior code readability and
minimalism in syntax so that programmers could successfully communicate their ideas in fewer lines of code.
1) Finding An Interpreter:
The first step towards implementing successful Python coding is finding an interpreter to execute your
programs.

Fortunately many online interpreters are available for immediate use without any prior downloads.

Alternatively Windows users have free access to several interpreters, such as the native IDLE (Integrated
Development Environment).

Dept of CSE, VVIET 2022-23 Page | 10

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

3.4 Hardware Requirements

CPU : Intel 2.1 GHZ

Memory: 4GB

Disk: 100GB

Display: 15 inch color

3.5 Software (Tools &Technologies) Requirements

Coding: Python
Platform: python
3.7 Tool :
Spyder
OS : Windows 7 Front end : tkinter in python

Dept of CSE, VVIET 2022-23 Page | 11

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Chapter 4
SYSTEM ANALYSIS
To tackle complex challenges effectively its essential to take an analytical approach that considers various
factors thoroughly. System analysis involves scrutinizing current problems in detail while also identifying
objectives and outlining potential solutions using both technical tools and critical thinking skills.
One fundamental component in this process is the feasibility study—an assessment that outlines specific goals
for design and development while also testing workability, resource efficiency, impact on organizational
performance among other factors. Our team has started conducting extensive research to complete a
comprehensive feasibility assessment based on prior research findings regarding our current issue at hand.

The eight crucial stages involved in carrying out an effective feasibility analysis are:

1) Establishing a project team with assigned leaders

2)Identifying potential systems proposed for implementation
3) Defining unique goals/features required from each potential system
4)Evaluating performance capabilities/cost effectiveness for every prospective solution Feasibility analysis
serves as a critical starting point when considering any new project or product development. It involves
examining three primary areas: economic feasibility (performance vs cost) technical feasibility (adaptability &
implementation) & social acceptability (public opinion). Economic viability forms an integral part of
considerations during this analysis phase, where we must carefully assess various performance metrics and
associated costs related to each proposed weight system.
By comparing all available options thoroughly and selecting the most suitable one for our projects goals and
resource allocation strategies- we can ensure maximum optimization of resources used while also meeting our
goals effectively.
With a final system selection made after extensively evaluating all options available under consideration during
feasibility analysis - it is time now for us to prepare an informative report.

Our report on selected recommendations for weight system implementation will provide key decision making
information necessary for management approval.

Dept of CSE, VVIET 2022-23 Page | 12

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

4.1.1 ECONOMICAL FEASIBILITY

This study is carried out to determine how the system will effect the company's bottom line. The company has a
restricted budget for research and development of the system. The justification for expenditures must be
provided. Because the majority of the technologies used are freely available, the constructed system was able to
remain below budget. Only customized products were required to be purchased.

4.1.2 TECHNICAL FEASIBILITY

The goal of this study is to look at the technological viability of the system, as well as its technical
requirements. Any new system must not need a big number of technical resources to function. As a
consequence, current technical resources will under significant strain. This will put the buyer under a lot of
stress. It is essential to make little or no changes to the developed system in order to implement this system.

4.1.3 SOCIAL FEASIBILITY

The study's purpose is to determine how effectively the system is accepted by its target audience. As part of
this, we train the user how to utilize the technology most effectively. The user should not be terrified of the
system, but rather accept it as a need. Only the methods employed to educate and acquaint the user with the
system have any effect on how well he adopts the system. Because he is the system's last user, it is critical that
his self-esteem be boosted so that he can offer some useful feedback.

Summary

The primary goal of this chapter is to determine whether or not the system is practical. So several types of
analysis, such as performance and technical, economic and financial analysis are carried out for these reasons.

Dept of CSE, VVIET 2022-23 Page | 13

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Chapter 5

SYSTEM DESIGN
A well-designed system is one that is easy to use and understand. For the sake of specifying a procedure or
system in sufficient depth to allow for its physical embodiment, "Design" is described as "The act of applying
numerous approaches and concepts." The system is built using a variety of design principles. System features,
components, and parts are all described in detail in the design specification for the benefit of end users.

5.1 Fundamental Design Concepts

Three decades ago, a set of core design ideals began to emerge. No matter how popular a certain idea may be at
any one moment, it has stood the test of time. One may build upon the other to create a more complex design
process from a solid base. In order to "get it right," designers rely on a set of core design ideas. There are
several essential design ideas that are used in this project to ensure that the specifications are met.

5.1.1 Input Design

User-oriented inputs are transformed into computer format via the process of input design. Input data should be
designed in such a way that it makes automation as simple and error-free as feasible. The application's user
interface is designed to make it simple for users to enter and choose data. When developing a project,
consideration is given to input design needs such as user friendliness, consistency in format and interaction
with users to ensure that they get exactly what they need and when they need it at all times. It is essential that
the input design be given the utmost care since it is an integral component of overall system design. One of the
most costly parts of a system is gathering input data since this data has to be routed via several modules. As a
result of this, if a user's IP address is unknown when they're ready to transmit the data to the destination, it
might lead to errors.

Dept of CSE, VVIET 2022-23 Page | 14

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

5.1.2 Design of Output

A high-quality output meets the demands of the end user while also successfully communicating the
information. In every system, outputs are utilized to communicate the results of processing to the user and other
systems. It provides the user with the most relevant and up-to- date information possible. The system's
interaction with the source and destination machines is improved through efficient and intelligent output.
Computer outputs are needed largely to ensure that the user receives the identical packet that they sent, rather
than damaged or faked packets. These findings are also utilized to save a copy of them for future reference.

5.1.3 The MVC Design Method

The model-delegate is a simplified version of the MVC concept that is used in Swing. In this architecture, both
the view and the controller object are combined into a single element called as the UI delegate. It is now
possible to converse back and forth between the model and the user interface delegate. Model and UI delegate
are the two components that make up a Swing component. The model is in charge of keeping track of the
current condition of each component. Each component's UI delegate is responsible for ensuring that the
necessary information is available for how the component should be shown on the computer screen. A variety
of events propagate across the component, and the UI delegate responds to them (along with AWT).

Fig 5.1- View and Controller are combined to form a UI delegate object.

Dept of CSE, VVIET 2022-23 Page | 15

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

The architecture of the system was designed using the MVC design pattern. As the foundation for each of its
components, the model-view-controller (MVC) architecture is used by the Swing framework. A GUI
component is broken into three parts by MVC. The component's behavior is heavily influenced by each of these
factors. A model, a view, and a controller are all parts of an MVC-style software application.

Model

The component's state and low-level behavior are represented by the model. It is in charge of maintaining the
current state and carrying out any changes on it. Neither the controllers nor the views are known to the model.
Each component's status data is included. There are a variety of models to choose from depending on the sort of
component you're dealing with. Information about the current location of the "thumb," its minimum and
maximum values and its width might be included in a scrollbar component's model. A menu, on the other hand,
may only show the options from which the user may choose. When the model's state changes, the system
informs the views and keeps track of the linkages between the model and the view.

View

An element's "view" describes how you see it displayed on the screen. The model's visual representation is
handled by this component. A title bar usually runs the full height of the window frame in the vast majority of
cases. Close boxes may be on either the left or right side of the title bar. The following are instances of distinct
views for the same window object.. Models may have several views, however in the Swing set, this is
uncommon.

Dept of CSE, VVIET 2022-23 Page | 16

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Fig 5.2- The MVC design allows for communication.

Controller

The model's user interface is controlled by the controller. The model's state may be modified using this
approach. A component's interaction with events is governed by its user interface.
The view will not be able to properly draw the scrollbar until it first obtains data from the model. Unless the
scrollbar can determine its current location and width in relation to the minimum and maximum values, the
"thumb" cannot be shown. It is determined by the view as to whether or not user events, including mouse clicks
are received by components. It's the controller's decision whether or not to act on these occurrences. The values
in the model may have to be changed based on the controller's choice. As a result of scrolling, the user's thumb
position in the model will be incremented. The cycle may be repeated at this point.

It is possible to divide the JFC user interface into three parts: the model, the view, and the controller. The basic
MVC paradigm is often adapted by combining the view and controller into a single unit. They're responsible
for the component's user-interface design.

Dept of CSE, VVIET 2022-23 Page | 17

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Figure 5.3- JFC user interface component

Dept of CSE, VVIET 2022-23 Page | 18

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

5.2 Methodology for System Development

In order to produce a product or solve an issue, the system development technique must be followed. There are
many stages, techniques, and actions involved in creating a piece of software. For product development, it
follows a set of stages. The waterfall model is used in this project's development.

5.2.1 Model phases

Requirement Analysis: This phase focuses on gathering the system's requirements. Documentation and
requirements review are part of this step.

System Design: The system specs are converted into a software representation with the needs in mind.
Algorithms, data structures, and software architecture are some of the topics covered in this phase.

Coding: In this phase, the programmer begins coding in order to sketch out the product's features in complete
detail. Machine-readable code is everything that is generated from system requirements.

Implementation: Software coding and programming are both part of the implementation process. Typically, the
product of this phase includes the library, executables, user guides, and other documentation for the program.

Testing: In this step, each program (model) is put together and tested to verify that the software requirements are
met. Verification and validation are the primary goals of testing.

Maintenance: Customers' needs vary over time; external environments change; faults and oversights discovered
during testing are fixed; and software efficiency is improved throughout the maintenance phase of development.

Dept of CSE, VVIET 2022-23 Page | 19

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

5.2.2 Why did you choose the waterfall model as your development method?

 “Clear project objectives.

 Stable project requirements.
 Progress of system is measurable.
 Strict sign-off requirements.
 Helps you to be perfect.
 Logic of software development is clearly understood.
 Production of a formal specification
 Better resource allocation.
 Improves quality. The emphasis on requirements and design before writing a single line of code
ensures minimal wastage of time and effort and reduces the risk of schedule slippage.
 Less human resources required as once one phase is finished those people can start working on to
the next phase.

Dept of CSE, VVIET 2022-23 Page | 20

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Fig 5.4:- The waterfall model

Architecture of the System

Conceptual design for a system is known as system architecture, which specifies the structure and behavior of
the system itself. A formal description of a system, arranged to facilitate reasoning about the structural aspects
of the system, is known as an architectural description. System components or building blocks are defined and
a strategy is provided for procuring and developing items and systems that operate together to achieve the
overall system.
The System architecture is shown below.

Dept of CSE, VVIET 2022-23 Page | 21

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Load Images Input Image

Train Using CNN

Classify
Load Image

Apply Watershed

Classified the
Image as Tobacco Plant or not

Fig 5.5:- The System Architecture

Dept of CSE, VVIET 2022-23 Page | 22

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Classes created specifically for the system

Using the Unified Modeling Language (UML), a class diagram is a form of static structural diagram that
explains the structure of a system by depicting the system's classes, their properties, and their interactions.
Below is a diagram of the classes.

Main

+LoadDataset()
+applySegmentUsingWatersh 1..*
ed() 1
+Training() Training
+Classification()
+loadDataset
1 ()
+ApplyCNN()
1..1

Classification

+loadInputImage()
+classifyTobaccoOrNotUsingCN
N()

Fig 5.6:- The Class Diagram

Dept of CSE, VVIET 2022-23 Page | 23

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

The system's use case diagram

Diagrams developed from a use case study are called use case diagrams. Use cases, actors, and any
relationships between them are all depicted graphically to provide a clear picture of the system's capabilities.

Collect Images

Train Dataset Using CNN

Segment the Image Using Watershed Algorithm

Classify Using CNN as Tobacco or Not

Fig 5.7:- The Use Case

Diagram Sequence diagram of system operation

Diagrams that indicate how processes interact with one another and in what order are known as sequence
diagrams in the Unified Modelling Language (UML). It's a chart's structure. The diagrams are shown in the
following order.

Dept of CSE, VVIET 2022-23 Page | 24

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

User RegionExtraction Watershed

LoadImage

ExtractRegions

Segment

segmentedImage

Fig 5.8:- The Sequence Diagram

User Training CNN Model

LoadDataset

TrainUsingCNN
ApplyCNN

Classify Tobacco or not

<ClassifiedResult>

Fig 5.9:- The Sequence Diagram

Dept of CSE, VVIET 2022-23 Page | 25

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Data Flow Diagram of the system

Graphical depiction of the "flow" of data is shown by the data-flow diagram. Data processing flow diagrams
(DFDs) are another use for DFDs (structured design). On a data flow diagram, data objects move from one
location to another through an internal process, either from an external data source or an internal data
repository.

Data flow diagram at Level 0

It shows the relationship between the system and its environment, which includes both data sources and sinks.
By concentrating simply on data flows across the system boundary, the context diagram (also known as Level 0
DFD) displays interactions between the system and its environment. If you just look at the context diagram,
you have no idea how the system is arranged.

Classified Result (tobacco or not)

Fig 5.10:- The Data flow diagram at Level 0

Level 1 Data flow diagram

Each sub-system (process) deals with one or more of the data flows to or from an external agent, and the
system as a whole is provided by these sub-systems (processes). An internal data storage is also identified, as
well as the movement of data from one element of a system to another, in this diagram.

Dept of CSE, VVIET 2022-23 Page | 26

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Load Input Images

Extract Region 1.0.1 Segmentation Using
Watershed1.0.2

Segmented Image

Load Dataset
Load Images
1.0.3
1.0.4

Classify Using CNN 1.0.6

Classified Result (tobacco or not) Apply CNN
1.0.5

Fig 5.11:- The Data flow diagram at Level 1

Summary
Most of this chapter is devoted to diagramming and explaining various aspects of how a
system works.

Dept Of CSE, VVIET 2022-23 Page | 27

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Chapter 6

IMPLEMENTATION
At this point in the project's lifecycle, the theoretical design has been translated into a
functioning system. The user department now bears the bulk of the responsibility for the
project's success or failure. Implementation without proper planning and management might
result in chaos and confusion.
The following are the tasks that must be completed during the implementation phase.

• “Careful planning.
• Investigation of system and constraints.
• Design of methods to achieve the changeover.
• Evaluation of the changeover method.
• Correct decisions regarding selection of the platform
• Appropriate selection of the language for application development”
.
6.1 The implementation language

As a result of this process, the final and proper product must be developed in an appropriate
programming language. Choosing the wrong programming language to develop a product
might lead to its demise.
Python is the programming language of choice for this project's implementation. Python's
popularity as a programming language may be summed up in the following few reasons: -

Platform Independence: Instead of generating native object code for a specific platform,
Python compilers generate instructions for the Python interpreter known as "byte code."
Writing a byte code interpreter to emulate a Spyder is all it takes to make Python programs
run on a certain platform. Basically, this implies every platform that supports Python may use
the same compiled byte code.

Dept Of CSE, VVIET 2022-23 Page | 28

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Objects Orientation: PYTHON IS A COMPLETE ORGANIZATIONAL LANGUAGE

Objects are the building blocks of all Python programs, therefore everything in a program is
descended from a root class of objects.

Rich Standard Library: Python's built-in library is one of the most appealing aspects of the
language. In addition to hundreds of classes and methods, the Python environment has six
core functional areas: -
“Language Support classes for advanced language features such as strings, arrays,
threads, and exception handling.
Utility classes like a random number generator, date and time functions, and container
classes.
Input/output classes to read and write data of many types to and from a variety of
sources.
Networking classes to allow inter-computer communications over a local network or
the Internet.
Abstract Window Toolkit for creating platform-independent GUI applications.
Applet is a class that lets you create Python programs that can be downloaded and run
on a client browser.”

Front end Interface: Python programmers may design standalone apps as well as web
applications that operate in a client browser after downloading from a web page.

Familiar C++-like Syntax: As a result of the closeness between Python syntax and C++
syntax, Python has become a popular programming language.

Garbage Collection: Memory allocated dynamically in Python does not need to be freed
directly by developers. Thus, Python programs are simpler to create and less likely to have
memory issues as a result of this improvement.

Dept Of CSE, VVIET 2022-23 Page | 29

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

6.2 Methodology

1. Load the image

2. To aid the thresholding step
a. Apply pyramid mean shift filtering to help the accuracy of our thresholding step, and
b. Finally display our image
3. Then convert the image to grayscale and apply Otsu’s thresholding to segment the
background
from the foreground.

4. Finally, the last step is to detect contours in the thresholder image and draw each
individual contour:
CNN Algorithm

Step 1: Choose a Dataset

Step 2: Prepare Dataset for Training Step 3: Create Training Data

Step 4: Assigning Labels and Features Step 5: Split X and Y for use in CNN
Step 6: Define, compile and train the CNN Model.

Dept Of CSE, VVIET 2022-23 Page | 30

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Chapter 7

TESTING
To fully test a PC-based framework, testers build a variety of tests to mimic diverse
scenarios. Every test has a distinct purpose, but their overall goal is the same: to make sure
that all of the framework's components are working together as intended and that they are
capable of their stated goals. The item is thoroughly tested to ensure that it performs exactly
as it should. Testing is the last stage of the association's testing and approval process.
During the testing phase, the following objectives are pursued:

To reaffirm the project's high quality.

To identify and correct any issues that may have lingered from the previous steps.
To demonstrate that the program is a viable solution to the original issue.
To guarantee the system's smooth functioning.
The primary focus of testing is on examining and making changes to the source code.

7.1 Unit Testing

Here, each component of the system is examined separately. Each module's smallest unit of
software design may be verified via unit testing. The framework's modules are tested one at a
time. Using the programming language itself, this testing is carried by Testing a module's
control structure for entire scope and maximum blunder location is done in certain methods
during unit testing.

7.2 Integration

Unit or module testing is completed before functions are merged into classes. Again, various
classes are integrated, and lastly, the front end and back end are integrated.

Function integration into classes

During the early stages of the coding process, just the functions that are absolutely necessary
to the program are created. Each function is written and tested on its own. The functions are

Dept Of CSE, VVIET 2022-23 Page | 31

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

included into their respective classes once they have been tested for correctness.

Integration of different classes

Here, each class is checked on its own to see whether it works properly. After each class has
been assessed individually, the results are combined and retested.

Integration of front-end with back-end

A Python Swing environment is used to create the front-end of the project. Using the user
interface, the user may enter instructions into the system and see how the system behaves in
both normal and abnormal situations. The GUI and back-end code are then tested together.

7.3 Integration Testing

The process of building a program's structure is called integration testing. It deals with the
complications that arise while trying to verify and build a program at the same time. To
construct a program from a set of unit tested modules, this testing procedure's primary goal is
to build on the design's specified program structure.
An extensive series of tests are carried out when the program is incorporated. Every module
is tested as a whole before it is released. The isolation of mistakes in this case is challenging
because of the sheer size of the software.

7.3.1 Top-down Integration

Building a program's structure in this way is a gradual process. Modules are integrated one at
a time, starting with the program's primary module and working their way down.

7.3.2. Bottom-up Inclusion

The following integration testing table shows the functions that were combined into
numerous classes and tested for functionality as a whole. For error-free communication
between classes and the integrity of stored data, this is essential.

Dept Of CSE, VVIET 2022-23 Page | 32

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

“Classes
integrated Functions integrated in each class Tests done Remarks
Class: Main Load Dataset () Class tested to Success
Training() check whether all
Classification() commands that
were appliedare
working correctly or
not.
Class: Training LoadDataset() Class tested to Success
buildCNNModel() check whether all
commands that
were appliedare
working correctly or
not.
Class: loadImage() Class tested to Success”
Classification WatershedSegmentation() check whether all
ClassifyImageAsTobacco commands that
OrNot() were appliedare
working correctly or
not.
Table 7.1:- Integration testing table

7.4 Validation Testing

Programming is completed and collected as a single unit at the end of combination testing.
The failures of the interface are exposed and corrected. There are several ways to define
approval testing. Tests confirm the product's capability in this case, as predicted by the
customer.

“Functionality to be
tested Input Tests done Remarks
Working of Front-EndUser interaction with help of a mouse and keyboard Appropriate forms
open when buttons are clicked Success
Working of Training module User must upload Image Dataset and do
Training Preprocessing done and load to build the CNN model
Success”
Working of Classification Load test Image and segment and classify with CNN
Trained Model Classify the image as Tobacco or Not. Success

Dept Of CSE, VVIET 2022-23 Page | 33

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

7.5 Output Testing

No system can be usable unless it can create the output in a certain format, hence the next
phase is output testing, which follows validation testing. That is why it's important to first
have a clear idea of what format the user needs and then evaluate the output provided by the
system in question. There are two ways to look at the output format: –
• On screen

• Printed format

7.6 User Acceptance Testing

The success of any system depends on how well the system is accepted by the people who
use it. Acceptance tests are really a performance by the end user. The success of the system
depends on the motivation and knowledge of the users.
At every stage of development, improvements are made to the system to ensure user
approval, particularly in reference to the following point: a regular dialogue with potential
system users. Design of the Input Screen
Design of the Output Display
A menu-based computer
system

7.6.1 Black box testing

The functional requirements of the program are the primary focus of black box testing.
Functional testing is another name for it. Using this method, the tester is not privy to the
internal workings of the program being tested. The programmer's code is never seen by the
tester, and the tester doesn't need to know anything about the program except for the
requirements. As a matter of fact, it is a supplementary method that is likely to identify faults
in the following areas:-
• The function is incorrect or missing.
• Errors with the interface.

Dept Of CSE, VVIET 2022-23 Page | 34

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

• Errors in performance
• Errors in initialization and termination
• Object mistakes

Advantages

• The test is impartial since the designer and tester are unrelated.
• The tester does not need to be familiar with any programming languages.
• The test is conducted from the perspective of the user, not the designer.
• As soon as the specs are finalized, test cases may be developed.

7.7 Test Data Preparation

When a system is being tested, it is essential that test data be prepared. Testing begins when a
set of test data has been gathered and prepared. Mistakes are identified and repaired during
testing by employing test data; these errors are also logged for future usage.

7.7.1 Using Artificial Test Data

In order to evaluate all possible formats and values, artificial test data are developed
especially for that purpose. In other words, a utility application in the information systems
department may rapidly generate fake data that can be used to test all of the program's login
and control pathways.

7.8 Quality Assurance

Management's auditing and reporting duties are part of quality assurance. With the help of
quality assurance, management may obtain a better understanding of product quality and feel
more confidence that it is fulfilling its objectives. In engineering, this is referred to as a
"umbrella activity." Quality assurance involves the following areas:-
Methods and tools for analysis, design, coding, and testing

The use of formal technical evaluations throughout each stage of the software
engineering process
The ability to keep tabs on the status of software documentation and the changes that
have been made.
A method for ensuring that software is developed in accordance with industry
standards.

Mechanisms for measuring and reporting.

Dept Of CSE, VVIET 2022-23 Page | 35

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

7.8.1 Quality Factors

The tracking of software quality and the evaluation of the effect of methodological and
procedural changes on enhanced software quality are major goals of quality assurance. There
are two basic categories of things that influence the quality:
Factors that can be directly measured.

Factors that can be indirectly measured

These factors focus on three important aspects of a software product

Characteristics of its use

The adaptability of the system
Its capacity to adapt to a new setting.
Performance in accomplishing its goals
how long the consumer will be able to use it.

7.8.2 Generic Risks

When anything goes wrong, it's called a risk. It is possible to separate project risks from other
occurrences by focusing on three factors:
A loss associated with the event.
The likelihood that the event will occur.
The degree to which we can change the outcome

7.8.3 Security Technologies & Policies

Various tasks connected with seven main activities make up software quality assurance:-
Application of technical methods.
Conduct of formal technical
reviews Software testing
Enforcement of standards
Control of change
Measurement
Record keeping and reporting

Dept Of CSE, VVIET 2022-23 Page | 36

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Summary
This chapter discusses many types of testing, such as unit testing, which is a technique for
verifying that a certain module of source code operates as expected. Module testing is another
name for this process.

Dept Of CSE, VVIET 2022-23 Page | 37

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Snapshots

Dept Of CSE, VVIET 2022-23 Page | 38

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Dept Of CSE, VVIET 2022-23 Page | 39

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Dept Of CSE, VVIET 2022-23 Page | 40

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Dept Of CSE, VVIET 2022-23 Page | 41

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Dept Of CSE, VVIET 2022-23 Page | 42

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

Chapter 8

CONCLUSION
Effective management of plant species in agriculture requires reliable plant recognition
capabilities, while botanists often use such applications for studying medicinal purposes. In
our project, classification refers to assigning a specific plant species to an image by analyzing
extracted features. Essentially, classification is accomplished by utilizing prior knowledge to
identify class labels associated with new input images. Our system aims at accurately
recognizing various types of plants in agricultural settings with an emphasis on speed and
accuracy during detection processes. Future efforts will concentrate on developing advanced
algorithms that deliver quick and precise results during the detection phase.
After considering all relevant techniques and methods discussed thus far, it's evident there are
multiple ways one can detect different types of plants; each method carries its own
advantages as well as limitations - highlighting opportunities for enhancing current practices
through further research. Amongst various approaches explored thus far during our research
journey, image processing stands out as particularly promising due to its ability in improving
existing research outcomes while generating fast and dependable results. Individuals who
undergo post secondary education at universities are equipped with aptitudes that do not just
benefit themselves but also create a ripple effect for enhanced well being across communities.
Through analytical thinking, research capabilities, and efficient communication university
educated individuals can address intricate issues faced by people from diverse backgrounds.
It is clear then that institutions of higher learning serve as key facilitators in developing
skilled professionals who offer significant contributions towards societal progress.

Dept Of CSE, VVIET 2022-23 Page | 43

Tobacco Plant Region Extraction and Segmentation using Watershed and CNN Algorithm

BIBILOGRAPHY
[1] L. Meng and J. P. Kerekes, “Object tracking using high resolution satellite imagery,”
IEEE J. Select. Topics Appl. Earth Observ. Remote Sens., vol. 5, no. 1, pp. 146–
152, Feb. 2012.
[2] JIANG HUIXIAN, The Analysis of Plants Image Recognition Based
on Deep Learning and Artificial Neural Network, Received March 20, 2020, accepted April
7, 2020, date of publication April 9, 2020, date of current version April 23, 2020
[3] D. R. Rossatto, D. Casanova, R. M. Kolb, and O. M. Bruno, ‘‘Fractal analysis of leaf-
texture properties as a tool for taxonomic and identification purposes: A case study with
species from neotropical melastomataceae (Miconieae tribe),’’ Plant Syst. Evol., vol. 291,
nos. 1–2, pp. 103– 116, Jan. 2011.
[4] E. Aptoula and B. Yanikoglu, ‘‘Morphological features for leaf based plant
recognition,’’ in
Proc. IEEE Int. Conf. Image Process., Sep. 2013, pp. 1496–1499.

[5] T. Ojala, M. Pietikäinen, and D. Harwood, ‘‘A comparative study of texture measures
with classification based on featured distributions,’’ Pattern Recognit., vol. 29, no. 1, pp. 51–
59, Jan. 1996.
[6] J. Chaki, R. Parekh, and S. Bhattacharya, ‘‘Plant leaf recognition using texture and
shape features with neural classifiers,’’ Pattern Recognit. Lett., vol. 58, pp. 61–68, Jun. 2015.
[7] K. Simonyan and A. Zisserman, ‘‘Very deep convolutional networks for large-scale
image
recognition,’’ in Proc. ICLR, Apr. 2015, pp. 1–14.

[8] M. J. Swain and D. H. Ballard, ‘‘Color indexing,’’ Int. J. Comput. Vis., vol. 7, no. 1,
pp. 11– 32, 1991.
[9] J. Yang and S. Zhu, ‘‘An online image retrieval method based on self color
correlogram model,’’ J. Comput. Inf. Syst., vol. 8, no. 8, pp. 3369–3376, 2012.
[10] J.-M. Guo, H. Prasetyo, and H.-S. Su, ‘‘Image indexing using the color and bit pattern
feature fusion,’’ J. Vis. Commun. Image Represent., vol. 24, no. 8, pp. 1360–1379, Nov.
2013.

Dept Of CSE, VVIET 2022-23 Page | 44