Raspberry Pi Image

Processing
Programming
With NumPy, SciPy, Matplotlib,
and OpenCV
Second Edition

Ashwin Pajankar
Raspberry Pi Image Processing Programming
Ashwin Pajankar
Nashik, Maharashtra, India

ISBN-13 (pbk): 978-1-4842-8269-4 ISBN-13 (electronic): 978-1-4842-8270-0

https://doi.org/10.1007/978-1-4842-8270-0

Copyright © 2022 by Ashwin Pajankar

This work is subject to copyright. All rights are reserved by the publisher, whether the whole or
part of the material is concerned, specifically the rights of translation, reprinting, reuse of
illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way,
and transmission or information storage and retrieval, electronic adaptation, computer software,
or by similar or dissimilar methodology now known or hereafter developed.
Trademarked names, logos, and images may appear in this book. Rather than use a trademark
symbol with every occurrence of a trademarked name, logo, or image we use the names, logos,
and images only in an editorial fashion and to the benefit of the trademark owner, with no
intention of infringement of the trademark.
The use in this publication of trade names, trademarks, service marks, and similar terms, even if
they are not identified as such, is not to be taken as an expression of opinion as to whether or not
they are subject to proprietary rights.
While the advice and information in this book are believed to be true and accurate at the date of
publication, neither the authors nor the editors nor the publisher can accept any legal
responsibility for any errors or omissions that may be made. The publisher makes no warranty,
express or implied, with respect to the material contained herein.
Managing Director, Apress Media LLC: Welmoed Spahr
Acquisitions Editor: Celestin Suresh John
Development Editor: James Markham
Coordinating Editor: Aditee Mirashi
Copyeditor: April Rondeau
Cover designed by eStudioCalamar
Cover image designed by Freepik (www.freepik.com)
Distributed to the book trade worldwide by Springer Science+Business Media New York, 1
New York Plaza, Suite 4600, New York, NY 10004-1562, USA. Phone 1-800-SPRINGER, fax (201)
348-4505, email [email protected], or visit www.springeronline.com. Apress Media,
LLC is a California LLC and the sole member (owner) is Springer Science+Business Media
Finance Inc (SSBM Finance Inc). SSBM Finance Inc is a Delaware corporation.
For information on translations, please e-mail [email protected]; for
reprint, paperback, or audio rights, please e-mail [email protected].
Apress titles may be purchased in bulk for academic, corporate, or promotional use. eBook
versions and licenses are also available for most titles. For more information, reference our Print
and eBook Bulk Sales web page at http://www.apress.com/bulk-sales.
Any source code or other supplementary material referenced by the author in this book is
available to readers on GitHub (github.com/apress) For more detailed information, please visit
http://www.apress.com/source-­code.
Printed on acid-free paper
 This book is dedicated to the memory of
Prof. Govindarajulu Regeti
(July 9, 1945 to March 18, 2021)

Popularly known to everyone as RGR, Prof. Govindarajulu

obtained a Bachelor of Technology degree in electrical
and electronic engineering from JNTU Kakinada. He also
earned a master’s degree and Ph.D. from IIT Kanpur.
Prof. Govindarajulu was an early faculty member of IIIT
Hyderabad and played a significant role in making IIIT
Hyderabad the top-class institution that it is today. He was
by far the most loved and cheered for faculty member of
the institute. He was full of energy to teach and full of old-­
fashioned charm. There is no doubt he cared for every student
as an individual, taking care to know about and to guide
them. He has taught, guided, and mentored many batches of
students at IIIT Hyderabad (including the author of this book).
Table of Contents
About the Author���������������������������������������������������������������������������������xi
About the Technical Reviewer�����������������������������������������������������������xiii
Acknowledgments������������������������������������������������������������������������������xv
Introduction��������������������������������������������������������������������������������������xvii

Chapter 1: Introduction to Single-Board Computers and

Raspberry Pi�����������������������������������������������������������������������������������������1
Single-Board Computers (SBCs)���������������������������������������������������������������������������1
Differences Between SBCs and Regular Computers���������������������������������������3
System on Chips (SoCs)�����������������������������������������������������������������������������������3
SoC versus Regular CPU����������������������������������������������������������������������������������4
History of SBCs������������������������������������������������������������������������������������������������4
Popular SBC Families��������������������������������������������������������������������������������������5
The Raspberry Pi���������������������������������������������������������������������������������������������������5
Raspberry Pi Setup���������������������������������������������������������������������������������������������10
Prepare the microSD Card�����������������������������������������������������������������������������12
Booting Up Raspberry Pi��������������������������������������������������������������������������������20
Accessing Raspberry Pi Remotely�����������������������������������������������������������������20
Configuring Raspberry Pi�������������������������������������������������������������������������������25
The Raspberry Pi OS�������������������������������������������������������������������������������������������27
The config.txt File������������������������������������������������������������������������������������������27
Updating the RPi��������������������������������������������������������������������������������������������28
Summary������������������������������������������������������������������������������������������������������������30

v
Table of Contents

Chapter 2: Introduction to Python and Digital Image Processing������31

A Brief History of the Python Programming Language���������������������������������������32
Features of Python����������������������������������������������������������������������������������������������34
Python 3��������������������������������������������������������������������������������������������������������������37
Python 2 and Python 3 on Raspberry Pi OS���������������������������������������������������38
Running a Python Program and Python Modes���������������������������������������������38
IDEs for Python����������������������������������������������������������������������������������������������������40
IDLE���������������������������������������������������������������������������������������������������������������40
Geany�������������������������������������������������������������������������������������������������������������43
Thonny IDE�����������������������������������������������������������������������������������������������������45
Introduction to Digital Image Processing������������������������������������������������������������46
Signal Processing������������������������������������������������������������������������������������������46
Image Processing������������������������������������������������������������������������������������������47
Using Raspberry Pi and Python for Digital Image Processing (DIP)���������������49
Exercise��������������������������������������������������������������������������������������������������������������49
Summary������������������������������������������������������������������������������������������������������������50

Chapter 3: Getting Started������������������������������������������������������������������51

Image Sources����������������������������������������������������������������������������������������������������51
Using a Webcam�������������������������������������������������������������������������������������������������54
guvcview�������������������������������������������������������������������������������������������������������55
fswebcam������������������������������������������������������������������������������������������������������56
The Pi Camera Module����������������������������������������������������������������������������������������57
raspistill���������������������������������������������������������������������������������������������������������61
Python 3 for Digital Image Processing����������������������������������������������������������������61
Working with Images�������������������������������������������������������������������������������������62
Summary������������������������������������������������������������������������������������������������������������67

vi
 Table of Contents

Chapter 4: Basic Operations on Images���������������������������������������������69

Image Module�����������������������������������������������������������������������������������������������������69
Image Channels���������������������������������������������������������������������������������������������70
Colorspace Conversion����������������������������������������������������������������������������������71
Image Blending����������������������������������������������������������������������������������������������72
Resizing an Image�����������������������������������������������������������������������������������������75
Rotating an Image�����������������������������������������������������������������������������������������76
Crop and Paste Operations����������������������������������������������������������������������������77
Copying and Saving Images to a File�������������������������������������������������������������79
Knowing the Value of a Particular Pixel���������������������������������������������������������79
Mandelbrot Set����������������������������������������������������������������������������������������������79
Julia Set���������������������������������������������������������������������������������������������������������81
Noise and Gradients��������������������������������������������������������������������������������������84
ImageChops Module�������������������������������������������������������������������������������������������85
ImageOps������������������������������������������������������������������������������������������������������������89
Summary������������������������������������������������������������������������������������������������������������90

Chapter 5: Advanced Operations on Images���������������������������������������91

ImageFilter Module���������������������������������������������������������������������������������������������91
The ImageEnhance Module�������������������������������������������������������������������������������105
Color Quantization���������������������������������������������������������������������������������������������108
Histograms and Equalization����������������������������������������������������������������������������110
Histogram Equalization��������������������������������������������������������������������������������111
Summary����������������������������������������������������������������������������������������������������������112

vii
Table of Contents

Chapter 6: Introduction to the Scientific Python Ecosystem������������113

The Scientific Python Ecosystem����������������������������������������������������������������������114
Simple Examples����������������������������������������������������������������������������������������������115
Matplotlib����������������������������������������������������������������������������������������������������������118
Image Channels�������������������������������������������������������������������������������������������123
Conversion Between PIL Image Objects and NumPy ndarrays�������������������������126
Summary����������������������������������������������������������������������������������������������������������127

Chapter 7: Transformations and Measurements������������������������������129

Transformations������������������������������������������������������������������������������������������������129
Measurements��������������������������������������������������������������������������������������������������139
Summary����������������������������������������������������������������������������������������������������������143

Chapter 8: Filters������������������������������������������������������������������������������145
Kernels, Convolution, and Correlation���������������������������������������������������������������146
Low-Pass Filters�����������������������������������������������������������������������������������������������150
Blurring��������������������������������������������������������������������������������������������������������150
Noise Reduction������������������������������������������������������������������������������������������156
High-Pass Filters�����������������������������������������������������������������������������������������������161
Fourier Filters����������������������������������������������������������������������������������������������������165
Summary����������������������������������������������������������������������������������������������������������167

Chapter 9: Morphology, Thresholding, and Segmentation����������������169

Distance Transforms�����������������������������������������������������������������������������������������169
Morphology and Morphological Operations������������������������������������������������������171
Structuring Element������������������������������������������������������������������������������������172
Binary Morphological Operations����������������������������������������������������������������172
Grayscale Morphological Operations�����������������������������������������������������������178
Thresholding and Segmentation�����������������������������������������������������������������������182
Summary����������������������������������������������������������������������������������������������������������188

viii
 Table of Contents

Chapter 10: Video Processing�����������������������������������������������������������189

Introduction to OpenCV Library�������������������������������������������������������������������������189
Converting the Colorspace of Image�����������������������������������������������������������������194
Separating Color Channels in Live Webcam Stream�����������������������������������������196
More Operations on Live Webcam Stream��������������������������������������������������������198
Geometric Transformation���������������������������������������������������������������������������198
Convolution��������������������������������������������������������������������������������������������������200
Correlation���������������������������������������������������������������������������������������������������204
Filtering�������������������������������������������������������������������������������������������������������206
Morphological Operations���������������������������������������������������������������������������216
Summary����������������������������������������������������������������������������������������������������������220
Conclusion��������������������������������������������������������������������������������������������������������220

Appendix�������������������������������������������������������������������������������������������223
pgmagick Image Processing�����������������������������������������������������������������������������223
Connecting a Display����������������������������������������������������������������������������������������225
Using a VGA Display������������������������������������������������������������������������������������225
Booting Up After Connecting a Display��������������������������������������������������������226
Connecting to Ethernet/ Wired Network������������������������������������������������������������234
Remote Desktop with VNC��������������������������������������������������������������������������������235

Index�������������������������������������������������������������������������������������������������241

ix
About the Author
Ashwin Pajankar earned a master of technology in computer science
engineering from IIIT Hyderabad and has over 25 years of experience
in the area of programming. He started his journey in programming
and electronics at the tender age of seven with an MS-DOS computer
and BASIC programming language. He is now proficient in Assembly
programming, C, C++, Java, shell scripting, JavaScript, Go Programming,
HTML, and Python. His other technical expertise includes single-board
computers such as Raspberry Pi and Banana Pro, microcontroller boards
such as Arduino, and embedded boards such as BBC Micro Bit. He has
extensively worked on domains such as software/product testing, software
automation, databases, data analytics and visualization, computer vision,
and web development.
He is currently a freelance online instructor teaching programming
and electronics to more than 82,000 professionals. He also regularly
conducts live programming bootcamps for software professionals.
His growing YouTube channel has an audience of more than 11,000
subscribers. He has published more than 20 books on programming and
electronics with many international publishers and is writing more books
with Apress. He also regularly reviews books on the topics of programming
and electronics written by other authors.
Apart from his work in the area of technology, he is active in the
community as a leader and volunteer for many social causes. He has
won several awards at his university (IIIT Hyderabad) and also at past
workplaces for his leadership in community service for uplifting the
underpriviledged with education and skill-based training and employment

xi
About the Author

assistance. He has also participated in many industry institute linkage

programs and connected his past employers with his alma maters. During
the COVID-19 pandemic (which was continuing at the time of writing
of this book), he participated in and led many initiatives to distribute
essential supplies such as food, clothes, and medicine to the needy people
in his local community. He regularly conducts initiatives for assisting
homeless people with clothes, essential supplies, and medical care.

xii
About the Technical Reviewer
Lentin Joseph is an author, roboticist, and
robotics entrepreneur from India. He runs
a robotics software company called Qbotics
Labs in Kochi/Kerala. He has ten years of
experience in the robotics domain, primarily
in the Robot Operating System, OpenCV,
and PCL.
He has authored ten books on ROS, namely,
Learning Robotics Using Python, first and
second editions; Mastering ROS for Robotics
Programming, first and second editions;
ROS Robotics Projects, first and second
editions; ROS Programming: Building Powerful Robots; and Robot
Operating System (ROS) for Absolute Beginners first and second editions.
He is also co-editor of the book Autonomous Driving and Advanced
Driver-Assistance Systems (ADAS): Applications, Development, Legal Issues,
and Testing.
He obtained his masters in robotics and automation from India and has
also worked at the Robotics Institute, Carnegie Mellon University, USA. He
is also a TEDx speaker.

xiii
Acknowledgments
I would like to thank my friend Anuradha for encouraging me to share
my knowledge with the world. I would like to thank my longtime mentors
from Apress, Celestin and Aditee, for giving me an opportunity to share
my knowledge and experience with readers. I thank Mark Powers and
James Markham for helping me to shape this book as per Apress standards.
I am in debt to the technical reviewer, Lentin Joseph. I also thank Prof.
Govindrajulu Sir’s family, Srinivas (son) and Amy (daughter-in-law),
for allowing me to dedicate this book to his memory and sharing with
us Govindrajulu Sir’s biographical information and his photograph for
publication. I would also like to thank all the people associated with Apress
who have been instrumental in bringing this project to reality.

xv
Introduction
I want to keep this introduction short, concise, and precise. I have been
working with the domain of image processing for quite a while now. I
was introduced to Python more than 15 years ago. When I first worked
with image processing using Raspberry Pi, I found it a bit tedious to comb
through all the literature available as printed books, video tutorials, and
online tutorials, as most lacked a step-by-step yet comprehensive guide for
a beginner getting started. It was then that I resolved to write a book, and
then I published the first edition of this book. It has been almost five years
since the first edition was published and it needed a lot of updates.
This book is the fruit of my efforts and cumulative experience of
thousands of hours (apart from the ones spent writing the actual book)
spent going through technical documentation, watching training videos,
writing code with the help of different tools, debugging faulty code
snippets, posting questions, and participating in discussions on various
question-answer and technical forums online, and referring to various
code repositories for directions. I have written this edition of the book in
such a way that beginners will find it easy to understand the topics. This
book has hundreds of code examples and images (of outputs of code
execution and screenshots) so readers can understand each concept
perfectly. All the code examples are adequately explained.
The book begins with a general discussion of Raspberry Pi and
Python programming. It is followed by a discussion of concepts related
to the domain of image processing. Then, it explores the libraries Pillow
and TKinter. The following few chapters focus on the scientific Python
ecosystem and image processing with libraries such as NumPy, SciPy,
and Matplotlib. The last chapter discusses how OpenCV can be used for

xvii
Introduction

applying image-processing routines on a live webcam feed. The appendix

briefly discusses the pgmagik library, which can be used for generating
sample images. It also has plenty of tips and tricks for using a Raspberry Pi
board effectively.
As a final note, we started and finished working on this project at a very
turbulent time (early 2022) when much of the world was facing yet another
wave of fresh infections brought by new variants of the virus causing the
COVID pandemic. Everyone around us is still recovering from the impact
of the pandemic. This book offered me a sense of purpose. I hope that all
the readers enjoy reading this book as much as I enjoyed writing it.

xviii
CHAPTER 1

Introduction
to Single-Board
Computers
and Raspberry Pi
Let’s start this exciting journey of exploring the scientific domain of digital
image processing with Raspberry Pi. To begin, we must be comfortable
with the basics of single-board computers (SBCs) and the Raspberry Pi.
This chapter discusses the definition, history, and philosophy of SBCs. It
compares SBCs to regular computers, then moves toward the most popular
and bestselling SBC of all time, the Raspberry Pi. By the end of this chapter,
we will have adequate knowledge to independently set up our own
Raspberry Pi. This chapter aims to make us comfortable with the basic
concepts of SBCs and Raspberry Pi setup.

Single-Board Computers (SBCs)

A single-board computer (referred to as an SBC from now on) is a fully
functional computer system built around a single printed circuit board. An

© Ashwin Pajankar 2022 1

A. Pajankar, Raspberry Pi Image Processing Programming,
https://doi.org/10.1007/978-1-4842-8270-0_1
Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

SBC has a microprocessor(s), memory, input/output, and other features

required of a minimally functioning computer. Unlike with desktop
personal computers (PC), most SBCs do not have expansion slots for
peripheral functions or expansion. As all components—processor(s),
RAM, GPU, etc.—are integrated on a single printed circuit board (PCB), we
cannot upgrade an SBC.
Few SBCs are made to plug into a backplane for system expansion.
SBCs come in many varieties, sizes, shapes, form factors, and feature sets.
Due to advances in electronics and semiconductor technologies, most
SBC prices are very low. With a price of around $50 a piece, we have in
our hands a development tool suitable for new applications, hacking,
debugging, testing, hardware development, and automation systems.
SBCs are usually manufactured with the following form factors:

• Pico-ITX

• PXI

• Qseven

• VMEbus

• VPX

• VXI
• AdvancedTCA

• CompactPCI

• Embedded Compact Extended (ECX)

• Mini-ITX

• PC/104

• PICMG

2
 Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

 ifferences Between SBCs

D
and Regular Computers
Table 1-1 lists the differences between SBCs and regular computers.

Table 1-1.  Differences Between SBCs and Regular Computers

Single-Board Computer Regular Computer

Not modular Modular

Components cannot be upgraded or Components can be upgraded or
replaced replaced
A system on chip Not a system on chip
Has a small form factor Has a large form factor
Is portable Is mostly non-portable or semi-portable
Consumes less power Consumes more power
Cheaper than a regular computer Costs more than an SBC

System on Chips (SoCs)

SBCs are predominantly systems on chips (SoCs). A system on a chip (SoC)
is an integrated circuit (IC) that has all the components of a computer on a
single chip. SoCs are very common with mobile electronic devices because
of their low power consumption and versatility. SoCs are widely used in
mobile phones, SBCs, and embedded hardware. An SoC includes all the
hardware and software needed for its operation.

3
Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

SoC versus Regular CPU

The biggest advantage of using an SoC is its size. If we use a CPU, it’s very
hard to make a compact computer because of the number of individual
chips and other components that we need to arrange on a board. However,
when using SoCs, we can place complete application-specific computing
systems in smartphones and tablets, and still have plenty of space for
batteries, the antenna, and other add-ons required for remote telephony
and data communication.
Due to the very high level of integration and the compact size, an
SoC uses considerably less power than a regular CPU. This is a significant
advantage of SoCs when it comes to mobile and portable systems. Also,
reducing the number of chips by eliminating redundant ICs on a computer
board results in a compact board size.

History of SBCs
Dyna-Micro was the first true SBC. It was based on the Intel C8080A and
used Intel’s first EPROM, the C1702A. The Dyna-Micro was rebranded
and marketed by E&L Instruments of Derby, Connecticut, in 1976 as
the MMD-1 (Mini-Micro Designer 1). It became famous as the leading
example of a microcomputer. SBCs were very popular in the earlier days of
computing, as many home computers were actually SBCs. However, with
the rise of PCs, the popularity of SBCs declined. Since 2010, there has been
a resurgence in the popularity of SBCs due to their lower production costs.
Apart from the MMD-1, here are a few other popular historical SBCs:

• The BBC Micro was built around an MOS technology

6502A processor running at 2MHz.

• The Ferguson Big Board II was a Zilog Z80-based

computer running at 4MHz.

• The Nascom was another Zilog Z80-based computer.

4
 Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

Popular SBC Families

Based on their manufacturers and designers, SBCs are grouped into
families, models, and generations. Here are a few popular SBC families:

• Raspberry Pi by the Raspberry Pi Foundation

• Banana Pi and Banana Pro

• Intel Up Squared Kits

• CubieBoard

• BeagleBone

The Raspberry Pi
The Raspberry Pi is a family of credit card–sized SBCs developed in the United
Kingdom by the Raspberry Pi Foundation. The Raspberry Pi Foundation was
formed by Eben Upton in 2009. The aim in developing the Raspberry Pi was
to promote the teaching of basic computer science in schools and developing
countries by providing a low-cost computing platform.
Raspberry Pi Foundation’s Raspberry Pi was released in 2012. It was
a massive hit and sold over two million units in two years. Subsequently,
the Raspberry Pi Foundation released revised versions of the Raspberry Pi.
They also released other accessories for the Pi.
More information about the Raspberry Pi Foundation can be found on
their website at https://www.raspberrypi.org.
The product page for Raspberry Pi's current production models
and other accessories can be found at https://www.raspberrypi.org/
products.
I have written, executed, and tested all the code examples in this
book on Raspberry Pi 4 Model B units with 8GB RAM. Table 1-2 lists the
specifications of the Raspberry Pi 4 Model B.

5
Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

Table 1-2.  Specifications of the Raspberry Pi 4 Model B

Release Date June 2019

Architecture ARMv8
SoC broadcom BCM2711
CPU Quad core Cortex-A72 (ARM v8) 64-bit SoC @ 1.5GHz
GPU Broadcom VideoCore IV
Memory 2GB, 4GB or 8GB LPDDR4-3200 SDRAM (depending on model)
USB 2 USB 3.0 ports, 2 USB 2.0 ports
Video output 2 × micro-HDMI ports (up to 4kp60 supported)
2-lane MIPI DSI display port
On-board storage Micro SDHC slot
On-board network 2.4 GHz and 5.0 GHz IEEE 802.11ac wireless
Bluetooth 5.0, BLE
Gigabit Ethernet
Power source 3A 5V via MicroUSB

Figure 1-1 shows the front view of the Raspberry Pi 4 Model B.

6
 Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

Figure 1-1.  The front view of Raspberry Pi 4 Model B board1

The image clearly shows all the important components of the board.
It shows USB ports, Ethernet port, micro-HDMI ports, 3.5mm audio jack,
USB type C port for power, and connectors for CSI and DSI.
Figure 1-2 shows the top view of the board.

1
 Image provided by Laserlicht under a CC-by-SA 4.0 license (https://
creativecommons.org/licenses/by-sa/4.0/deed.en)

7
Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

Figure 1-2.  The top view of Raspberry Pi 4 Model B board2

We can see the microSD card slot clearly in the bottom view
(Figure 1-3).

2
 Image provided by Laserlicht under a CC-by-SA 4.0 license (https://
creativecommons.org/licenses/by-sa/4.0/deed.en)

8
 Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

Figure 1-3.  The bottom view of a Raspberry Pi 4 Model B board with

a microSD card

9
Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

Raspberry Pi Setup
We have to set up Raspberry Pi before we can use it for exploration and
adventure. As mentioned earlier, I am using a Raspberry Pi 4 Model B for
this setup. Make sure that you use a model of RPi board that has built-in
Wi-Fi. The setup process is almost the same for all models.
We are going to use the Raspberry Pi board in headless mode. This
means that we will not connect any keyboard, mouse, or display. We will
just connect it to the Wi-Fi network and access it remotely. So, we are not
going to need a lot of hardware for this. Apart from the RPi board itself, we
will need the following things:

A computer – We will need a computer with a

Windows, macOS, Linux, UNIX, or BSD operating
system. We need it for preparing the microSD card
for the RPi board.

Wi-Fi connection with internet – We need the

internet to install the operating system on the
microSD card. We will need it further to connect the
RPi to the internet and install necessary software on
the RPi.

microSD card – We will need a microSD card.

I usually recommend at least a 32GB Class 10
microSD card for the best experience. If you wish
to know more about the classes of microSD cards,
please visit the following URL on the web: https://
www.sdcard.org/developers/sd-standard-
overview/speed-class/
microSD card reader – We need this so that the
computer can read from and write to the microSD
card. We need to modify the contents of the

10
 Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

microSD card for headless setup. There are many

variations of microSD card reader. The following
(Figure 1-4) is an image of a universal card reader.

Figure 1-4.  Universal card reader3

A power supply – We need a power supply to power

the RPi board. RPi 4B uses a USB Type C power
supply like the one at https://www.raspberrypi.
com/products/type-c-power-supply/, and a few
other models use a micro USB power supply like the
one at https://www.raspberrypi.com/products/
micro-usb-power-supply/.

We can procure all these things and the RPi boards at online
marketplaces or local hobby electronics stores.

3
 Image provided by Alen under a CC0 1.0 Universal (CC0 1.0) license (https://
creativecommons.org/publicdomain/zero/1.0/)

11
Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

Prepare the microSD Card

Following are the steps to prepare the card:

1) Insert the card into the card reader and connect it to

the computer.

2) Install the Raspberry Pi Imager software from

https://www.raspberrypi.com/software/. For
Debian and derivatives, run the following command
in the terminal emulator:

sudo apt install rpi-imager

3) Open the installed Raspberry Pi Imager program. It

looks as shown in Figure 1-5.

Figure 1-5.  Raspberry Pi Imager

12
 Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

4) Click on the CHOOSE OS button. It opens options

as shown in Figure 1-6.

Figure 1-6.  Choosing the OS

Choose the option Raspberry Pi OS (other) by clicking on it. It shows

another list of options, as shown in Figure 1-7.

13
Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

Figure 1-7.  Choosing the OS

Choose Raspberry Pi OS (Legacy). It is a very well-maintained flavor

of Debian Buster that supports command-line utilities for the Raspberry
Pi camera.

5) Now, click on the CHOOSE STORAGE button. It will

show the microSD card connected to the computer,
as shown in Figure 1-8.

14
 Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

Figure 1-8.  Choosing the storage

6) It will enable the WRITE button, as shown in

Figure 1-9.

15
Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

Figure 1-9.  Write button enabled

Click the WRITE button. It will show a warning message, as

shown in Figure 1-10.

16
 Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

Figure 1-10.  Warning message

Don’t worry about the message. Click the YES button.

It starts downloading the files from the internet and
writing them to the microSD card. A progress bar will
appear, as shown in Figure 1-11.

17
Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

Figure 1-11.  Progress bar

Once the entire process completes, it shows the

following (Figure 1-12) message.

18
 Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

Figure 1-12.  Final message

Click the CONTINUE button. Disconnect the card

reader and reconnect it again.

7) Once we reconnect it, it will show in the filesystem

as a drive labeled boot. The OS may automatically
assign a letter to that. Create an empty file in the
location boot and name it ssh. Don’t use any
extension. This will enable remote login.

8) Create another file and name it wpa_supplicant.

conf. It is necessary to connect the RPi board to the
Wi-Fi after booting up. Add the following contents to
the file and save it:

country=IN
ctrl_interface=DIR=/var/run/wpa_supplicant
GROUP=netdev

19
Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

update_config=1

network={
scan_ssid=1
ssid="TP-Link_710E"
psk="internet1"
}

Remember to change the SSID and PSK sections

with the name and the password of your Wi-Fi.
9) Now, safely disconnect the microSD card reader.

Booting Up Raspberry Pi
This is the easiest part. Following these steps:

1) Remove the microSD card from the card reader and

insert it into the microSD card slot of the RPi board.

2) Make sure that the main power switch is in the

OFF position at this point. Connect the RPi to the
power supply.
3) Switch on the power switch.

We can see the lights on the RPi board blinking at this point. This
means that the RPi is booting up. Wait for a couple of minutes for the boot
process to complete.

Accessing Raspberry Pi Remotely

We have booted up the RPi board in headless mode. This means that we
have not directly connected any I/O devices such as display, keyboard, or
mouse to it. Let’s connect to it remotely. First, we must know its IP address.

20
 Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

Since we enabled the SSH and provided it the settings of our Wi-Fi before
booting up, it is connected to the Wi-Fi at the time of the booting process.
There are many ways to find out the IP address. If you are a part of an
organization (workplace, research lab, or university), check with your
network/system administrator to find out the IP address of the RPi board.
If it is your personal Wi-Fi, then on all the UNIX-like systems (Linux,
BSD, macOS), you can run a command to find out the IP address. For
Debian and derivatives (Ubuntu, Raspberry Pi OS, etc.), we can install the
following command:

sudo apt install nmap -y

This utility (NMAP) scans the network. The command to find out all
the IP addresses connected to the network is as follows:

sudo nmap -sn 192.168.0.*

Change the numbers to represent your own network. It produces the

following output:

Starting Nmap 7.80 ( https://nmap.org ) at 2022-02-01 11:26 IST

Nmap scan report for 192.168.0.1
Host is up (0.0025s latency).
Nmap scan report for 192.168.0.100
Host is up (0.0023s latency).
Nmap scan report for 192.168.0.102
Host is up (0.091s latency).
Nmap done: 256 IP addresses (3 hosts up) scanned in
2.72 seconds

You can find the IP address of your own system with the commands
ipconfig (Windows) or ifconfig (UNIX-like systems). From this list, we
can eliminate all the known devices with known IP addresses. If there are
too many devices attached to the home network, then turn off the Wi-Fi of

21
Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

all the unnecessary devices. In my case, the IP address of the RPi board is
192.168.0.100.
On a Windows computer, we can install the Zenmap utility, which is
the graphical interface for the nmap. We can download and install it from
https://nmap.org/zenmap/. It is also available for UNIX-like systems.
Install the utility and open it. It looks as shown in Figure 1-13.

Figure 1-13.  Zenmap GUI

We can easily use it. Mention 192.168.0.* in the Target text box.
Mention the command that we used earlier (nmap -sn 192.168.0.*) in
the Command text box. And then click the Scan button. It takes some
time for the scan to finish. Sometimes the scan fails prematurely. In such
cases, keep trying, and it will work after a couple of attempts. It is all open
source and free, so we cannot complain. It produces the result shown
in Figure 1-13 once finished. Sometimes, it also shows the names of the
manufacturers of the connected devices beside their MAC addresses. This
makes it easy to identify the RPi board.

22
 Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

Once we have the IP address of the RPi board, we can log in remotely.
We can use a variety of SSH tools, such as the built-in ssh utility on UNIX-
like operating systems, to remotely log in to the RPi board. Just run the
following command in the terminal emulator of your operating system:

ssh [email protected]

The pi is the username. It will prompt for the password. It is raspberry.

This way, we can remotely log in to the RPi board and run commands.
PuTTY and Bitvise SSH clients are a few popular ones on Windows.
However, I find MobaXterm very handy for this. It has the options for
remote terminal, file transfer, and X-11 forwarding. We do not have
to install anything else if we have this tool. Install it from https://
mobaxterm.mobatek.net/. Once installed, open it. The window looks as
shown in Figure 1-14.

Figure 1-14.  MobaXterm

23
Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

In this window, click on the Start local terminal button. It launches

a local terminal window. There, we can run the following command to
connect to the RPi board:

ssh [email protected]

The pi is the username. It will prompt for the password. It is raspberry.

This way, we can remotely log in to the RPi board and run commands.
Now, it is time to brush up on the UNIX concepts and commands
that we learned in school. We can run all the Linux/UNIX commands on
this remote terminal. One of the best things about the MobaXterm is it
has a built-in XServer program that is automatically launched when we
launch this utility. We can see the symbol for that in the top-right corner
of the window. UNIX-like systems allow us to access the GUI applications
remotely with X-11 forwarding. In RPi, it is enabled by default. We just
need an XServer utility at the other end (the computer from which we
are connecting to the RPi, or any other UNIX-like system). As discussed
already, we have it with the MobaXterm. Just run the following command
in the terminal:

nohup pcmanfm &

It opens a window for the File Explorer utility, as shown in Figure 1-15.

24
 Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

Figure 1-15.  File Explorer accessed with X-11 forwarding

Keep in mind that this File Explorer utility is running on the RPi.
We are accessing its GUI using X-11 Forwarding. This way, with the
combination of nohup (no hangup) and &, we can launch any GUI utility
from the terminal. & makes sure that the control of the terminal is returned
back to us, and nohup keeps the process running even when the user
logs out.

Configuring Raspberry Pi
Now, type sudo raspi-config in the prompt and press Enter. It opens
the tool raspi-config, which is the configuration tool for the Raspberry Pi
OS. First, update it, as shown in Figure 1-16.

25
Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

Figure 1-16.  Updating the raspi-config tool

It will take some time to update. Once updated, go to the fifth option
for Localization. It looks as follows (Figure 1-17).

Figure 1-17.  Localization

Set all these options as per your choice, and in the main menu choose
Finish. It will ask to reboot. Choose Yes, and it will reboot the RPi.

26
 Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

The Raspberry Pi OS
An operating system is the set of basic programs and utilities that make
a computer work. It is an interface between the user and the computer.
Raspberry Pi OS is a free operating system based on the popular Linux
distribution Debian. It is optimized for the Raspberry Pi family of SBCs. It
is even ported to the other, similar SBCs like Banana Pro.

The config.txt File

Raspberry Pi does not have a conventional BIOS or UEFI. The BIOS (basic
input/output system) is the program that a computer’s microprocessor
uses to get the computer system started after it is turned on. It also
manages data flow between the computer’s operating system and attached
peripheral devices, such as the hard disk, video adapter, keyboard, mouse,
and printer.
Since the Raspberry Pi does not have a BIOS/UEFI, the various system
configuration parameters that are normally stored and modified using the
BIOS/UEFI are instead stored in a text file called config.txt.
The Raspberry Pi config.txt file is on the boot partition of the Raspberry
Pi. It is normally accessible as /boot/config.txt from Linux. However,
from Windows and MacOS, it is seen as a file in the accessible part of
the microSD card. The accessible part of the card, as we already know, is
labeled boot.
On a Raspberry Pi, you can edit this file with the following command:

sudo nano /boot/config.txt

27
Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

Note  nano is a simple and easy-to-learn terminal-based text editor

for Linux. Visit its home page at https://www.nano-editor.org
to learn more about it. I find it easier to use than vi or vim editors.
To learn more about config.txt, visit the page http://elinux.org/
RPiconfig. A sample configuration can also be found at http://
elinux.org/R-Pi_configuration_file.

Updating the RPi
The Raspberry Pi must be connected to the internet in order to update it
successfully. Let’s update the firmware and the Raspberry Pi OS.

Updating the Firmware
To update the firmware, run the command sudo rpi-update.

Updating and Upgrading Raspberry Pi OS

We will use APT for this. APT (Advanced Package Tool) is a program that
handles the installation and removal of software on Debian and other
Debian derivatives. APT simplifies the process of managing software on
Debian systems by automating the fetching, configuration, and installation
of software packages. We need an internet connection for this too.
First, update the system’s package list by entering the following
command in the terminal:

sudo apt update

28
 Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

This downloads the package lists from the respective remote

repositories and updates them in the local computer so that information
on the newest versions of packages and their dependencies is available for
the installation and update. It should be run before running the install or
upgrade commands.
Next, upgrade all the installed packages to their latest versions using
this command:

sudo apt full-upgrade -y

This fetches new versions of the packages on the local machine that are
marked for upgrade. It also detects and installs any dependencies. It also
removes obsolete packages.
Doing this regularly will keep the Raspberry Pi OS up to date. After
entering these commands, it will take a while to update the OS, because
these commands fetch the data and the packages from remote repositories
on the internet.

Note The command apt help will list all the options associated
with the APT utility.

Shutting Down and Restarting RPi

We can shut down the RPi safely using either of the following commands:

sudo shutdown -h
sudo init 0

We can reboot the RPi using either of the following commands:

sudo reboot -h
sudo init 6

29
Chapter 1 Introduction to Single-Board Computers and Raspberry Pi

S
 ummary
In this chapter, we learned how to set up and access RPi in headless mode
over Wi-Fi. We are comfortable with this part now.
The next chapter will focus on the concepts involved with digital
image processing. We will also write basic programs with Python 3.

30
CHAPTER 2

Introduction
to Python and Digital
Image Processing
In the last chapter, we explored the amazing world of single-board
computers and Raspberry Pi. We booted up the Raspberry Pi, connected it
to the internet, and updated the Raspberry Pi OS.
In this chapter, we will get started with Python and the concepts
of digital image processing (DIP). Let’s begin with an introduction to
Python. I personally find Python amazing and have been enchanted
by it. Python is a simple yet powerful programming language. When
programmers use Python, it’s easy to focus on solving a given problem
as they do not have to worry about the syntax. Python perfectly fits the
philosophy of Raspberry Pi, which is programming for everyone. That’s
why it’s the most preferred programming platform for Raspberry Pi and
other computers.
The following is a list of topics we will learn in this chapter:

• History of Python

• Features of Python

• Python 3

© Ashwin Pajankar 2022 31

A. Pajankar, Raspberry Pi Image Processing Programming,
https://doi.org/10.1007/978-1-4842-8270-0_2
Chapter 2 Introduction to Python and Digital Image Processing

• IDEs for Python

• Introduction to digital image processing

By the end of this chapter, we will be comfortable with the concepts of

digital image processing and Python 3 basics.

 Brief History of the Python

A
Programming Language
Python was designed and conceived in the late 1980s. Its actual
implementation was started in late 1989 by Guido van Rossum at
Centrum Wiskunde & Informatica (National Research Institute for
Mathematics and Computer Science) in the Netherlands. Python is a
successor to the ABC programming language, which itself was inspired
by the SETL programming language. In February of 1991, Van Rossum
publicly published the Python source code to the newsgroup alt.sources.
The name Python was inspired by the British television show Monty
Python’s Flying Circus because Van Rossum was a big fan of Monty Python.
Van Rossum is the principal author of the Python programming
language. He played a central role in guiding the development,
enhancement, and further evolution of Python. He held the title
Benevolent Dictator for Life for Python till July 2018. Currently, Python
development is guided by a five-member steering council.
The central philosophy of the Python programming language (the Zen
of Python) is explained in PEP-20 (PEP stands for Python Enhancement
Proposal), which can be found at https://www.python.org/dev/peps/
pep-­0020.
It is a collection of 20 software principles, out of which 19 have been
documented. The principles are as follows:

• Beautiful is better than ugly.

• Explicit is better than implicit.

32
 Chapter 2 Introduction to Python and Digital Image Processing

• Simple is better than complex.

• Complex is better than complicated.

• Flat is better than nested.

• Sparse is better than dense.

• Readability counts.

• Special cases aren’t special enough to break the rules.

• Practicality beats purity.

• Errors should never pass silently.

• Unless explicitly silenced.

• In the face of ambiguity, refuse the temptation to guess.

• There should be one—and preferably only one—

obvious way to do it.

• Although that way may not be obvious at first unless

you’re Dutch.

• Now is better than never.

• Although never is often better than right now.

• If the implementation is hard to explain, it’s a bad idea.

• If the implementation is easy to explain, it may be a

good idea.

• Namespaces are one honking great idea—let’s do more

of those!

33
Chapter 2 Introduction to Python and Digital Image Processing

Features of Python
The following are the features of Python that have made it popular and
beloved in the programming community.
Simple
Python is a simple language with a minimalist approach. Reading a
well-written and good Python program makes you think you are reading
English text.
Easy to Learn
Due to its simple and English-like syntax, Python is extremely easy to
learn. That is the prime reason why it is taught as the first programming
language to high school and university students who take introductory
programming courses. An entire new generation of programmers is
learning Python as their first programming language.
Easy to Read
Unlike other high-level programming languages, Python does not
obfuscate the code and make it unreadable. The English-like structure of
the Python code makes it easier to read compared to code written in other
programming languages. This makes it easier to understand and easier to
learn compared to other high-level languages like C and C++.
Easy to Maintain
As Python code is easy to read, easy to understand, and easy to learn,
anyone maintaining the code becomes comfortable with the codebase
very quickly. I can vouch for this from my personal experience maintaining
and enhancing large legacy codebases that were written in a combination
of Bash and Python 2.
Open Source
Python is an open source project, which means its source code is
freely available. You can make changes to it to suit your needs and use the
original and modified code in your applications.

34
 Chapter 2 Introduction to Python and Digital Image Processing

High-Level Language
While writing Python programs, you do not have to manage the low-­
level details like memory management, CPU timings, and scheduling
processes. All these tasks are managed by the Python interpreter. You can
write the code directly in easy-to-understand English-like syntax.
Portable
The Python interpreter has been ported to many OS platforms. Python
code is also portable. All the Python programs will work on the supported
platform without requiring many changes if you are careful to avoid
system-dependent coding.
You can use Python on GNU/Linux, Windows, Android, FreeBSD, Mac
OS, iOS, Solaris, OS/2, Amiga, Palm OS, QNX, VMS, AROS, AS/400, BeOS,
OS/390, z/OS, Psion, Acorn, PlayStation, Sharp Zaurus, RISC OS, VxWorks,
Windows CE, and PocketPC.
Interpreted
Python is an interpreted language. Let’s take a look at what that means.
Programs written in high-level programming languages like C, C++, and
Java are compiled first. This means that they are first converted into an
intermediate format. When we run the program, this intermediate format
is loaded from secondary storage (i.e., from the hard disk) to the memory
(RAM) by the linker/loader.
So, C, C++, and Java have a separate compiler and linker/loader. This
is not the case with Python. Python runs the program directly from the
source code. You do not have to bother compiling and linking to the proper
libraries. This makes Python programs truly portable, as you can copy the
program to one computer from another and the program runs fine as long
as the necessary libraries are installed on the target computer.
Object Oriented
Python supports procedure-oriented programming as well as object-­
oriented programming paradigms.

35
Chapter 2 Introduction to Python and Digital Image Processing

All the object-oriented programming paradigms are implemented in

Python. In object-oriented programming languages, the program is built
around objects that combine data and related functionality. Python is a
very simple but powerful object-oriented programming language.
Extensible
One of the features of Python is that you can call C and C++ routines
from the Python programs. If you want the core functionality of the
application to run faster, you can code that part in C/C++ and call it in the
Python program (C/C++ programs generally run faster than Python ones).
Extensive Libraries
Python has an extensive standard library that comes pre-installed.
The standard library has all the essential features for a modern-day
programming language. It has provisions for databases, unit testing (we
will explore this later in this book), regular expressions, multi-threading,
network programming, computer graphics, image processing, GUI, and
other utilities. This is part of Python’s batteries-included philosophy.
Apart from the standard library, Python has numerous and ever-­
growing sets of third-party libraries. The list of these libraries can be found
on the Python Package Index.
Robust
Python provides robustness by means of the ability to handle errors.
The full stack trace of the encountered errors is available and makes the
life of the programmer more bearable. The runtime errors are known as
exceptions. The feature that handles these errors is known as an exception-­
handling mechanism.
Rapid Prototyping
Python is used as a rapid prototyping tool. As you learned earlier,
Python has extensive libraries and is easy to learn, which has led many
software architects to use it as a tool to rapidly prototype their ideas into
working models.

36
 Chapter 2 Introduction to Python and Digital Image Processing

Memory Management
In assembly language and in programming languages like C and C++,
memory management is the responsibility of the programmer. This is in
addition to the task at hand. This creates an unnecessary burden on the
programmer. In Python, the Python interpreter takes care of the memory
management. This helps the programmers steer clear of memory issues
and focus on the task at hand.
Powerful
Python has everything in it that a modern programming language
needs. It is used in applications such as computer visioning,
supercomputing, drug discovery, scientific computing, simulation, and
bioinformatics. Millions of programmers around the world use Python.
Many big organizations like NASA, Google, SpaceX, and Cisco use Python
for their applications and infrastructure.
Community Support
I find this to be the most appealing feature of Python. Recall that
Python is open source. It also has a community of almost a million
programmers throughout the world (probably more, as today high school
kids are learning Python too). There are also plenty of forums on the
internet to support programmers who encounter a roadblock. None of my
queries related to Python have ever gone unanswered.

Python 3
Python 3 was released in 2008. The Python development team decided
to do away with some of the redundant features of Python, simplify some
more features, rectify some design flaws, and add a few much-needed
features.
It was decided that a major revision number was needed for this and
that the resultant release would not be backward compatible. Python
2.x and 3.x were supposed to coexist in parallel so the programmer

37
Chapter 2 Introduction to Python and Digital Image Processing

community would have enough time to migrate their code and third-party
libraries from 2.x to 3.x. Python 2.x code cannot be run as-is in most cases,
as there are significant differences between 2.x and 3.x.

Python 2 and Python 3 on Raspberry Pi OS

Raspberry Pi OS is a derivative of Debian Linux. Python 2 and Python 3
interpreters are pre-installed in Raspberry Pi OS. The Python 2 interpreter
can be invoked by running the command python in the terminal. The
Python 3 interpreter can be invoked by running the command python3 in
the terminal. We can check the Python 3 interpreter version by running
either of the following commands:

python3 -V
python --version

You can check the location of the Python 3 binary by running the
following command:

which python3

Running a Python Program and Python Modes

You have set up your environment for Python programming, so let’s get
started with the simple concepts of Python. Python has two modes—
interactive mode and script mode. Let’s look at these modes in detail.

Interactive Mode
Python’s interactive mode is a command-line shell. It provides immediate
output for every executed statement. It also stores the output of previously
executed statements in the active memory. As new statements are
executed by the Python interpreter, the entire sequence of previously

38
 Chapter 2 Introduction to Python and Digital Image Processing

executed statements is considered while evaluating the current output. We

have to type the command python3 in the terminal to invoke the Python 3
interpreter into interactive mode, as follows:

pi@raspberrypi:~ $
Python 3.4.2 (default, Oct 19 2014, 13:31:11)
[GCC 4.9.1] on linux
Type "help", "copyright", "credits" or "license" for more
information.
>>>

You can execute Python statements directly in this interactive mode

just like when you run commands from the OS shell/console, as follows:

>>>print ('Hello World!')

Hello World!
>>>

We will not be using interactive mode in this book. However, keep it

in mind, as it’s the quickest way to check small snippets of code (five to
ten lines). We can quit interactive mode using the exit() statement, as
follows:

>>> exit()
pi@raspberrypi:~ $

Script Mode
Script mode is where the Python script files (.py) are executed by the
Python interpreter.

39
Chapter 2 Introduction to Python and Digital Image Processing

Create a file called test.py and add the print ('Hello World!')
statement to the file. Save the file and run it with the Python 3 interpreter
as follows:

pi@raspberrypi:~ $ python3 test.py

Hello World!
pi@raspberrypi:~ $

In this example, python3 is the interpreter and test.py is the filename.

In case the Python test.py file is not in the same directory where we are
invoking the python3 interpreter, we have to provide the absolute path of
the Python file.

IDEs for Python
An integrated development environment (IDE) is a software suite that has
all the basic tool s needed to write and test programs. A typical IDE has
a compiler, a debugger, a code editor, and a build automation tool. Most
programming languages have various IDEs to make programmers’ lives
better. Python too has many IDEs. Let’s look at a few of them.

IDLE
IDLE stands for Integrated DeveLopment Environment. We have to install
it on Raspberry Pi OS. IDLE3 is for Python 3. It’s popular with Python
beginners. We can install it with the following command:

pip3 install idle

We can launch IDLE3 using the following command:

nohup idle &

Figure 2-1 shows the IDLE3 code editor and an interactive prompt.

40
 Chapter 2 Introduction to Python and Digital Image Processing

Figure 2-1.  IDLE3 interactive prompt for Python 3 interpreter and a

new code file

We can go to Options ➤ Configure IDLE in the main menu. We can

change the settings of the editor here (Figure 2-2).

41
Chapter 2 Introduction to Python and Digital Image Processing

Figure 2-2.  IDLE3 configuration options

42
 Chapter 2 Introduction to Python and Digital Image Processing

Geany
Geany is a text editor that uses the GTK+ toolkit and has the basic features
of an integrated development environment. It supports many file types
and programming languages. It has some nice features. Check out
https://www.geany.org for more details. Figure 2-3 shows the Geany text
editor window.

Figure 2-3.  Geany IDE

Type print("Hello World!") in the code editor and save the file in
the /home/pi directory as test.py. Click Build in the menu bar and then
choose Execute. You can also use the F5 keyboard shortcut to execute
the program. The program will execute in a terminal window. You have to
press Enter to close the execution window. The default Python interpreter
for Geany is Python 2. We will need to change it to Python 3. To do that,
go to Build ➤ Set Build Commands. The window shown in Figure 2-4
will appear.

43
Chapter 2 Introduction to Python and Digital Image Processing

Figure 2-4.  Set Build Commands window

In the Set Build Commands window, within the Execute Commands

section, change python “%f” (highlighted in the box in Figure 2-4) to
python3“%f” to set the Python 3 interpreter as the default interpreter.
After that, run the program again to verify that everything works correctly.

44
 Chapter 2 Introduction to Python and Digital Image Processing

Thonny IDE
Thonny IDE also comes installed with the latest edition of the Raspberry
Pi OS. It works with Python 3 by default (and only with Python 3, so we
do not have to change the settings). We can invoke it with the following
command:

nohup thonny &

Figure 2-5 shows a screenshot of the Thonny IDE instance.

Figure 2-5.  Thonny IDE

45
Chapter 2 Introduction to Python and Digital Image Processing

Introduction to Digital Image Processing

To get started with digital image processing (DIP), we will briefly review
our understanding of a few basic concepts about related topics.

Signal Processing
Anything that carries any information, when represented mathematically,
is called a signal. The process or technique used to extract useful and
relevant information from a given signal is known as signal processing. The
system that does this type of task is known as a signal processing system.
The best example of a signal processing system is the human brain. It
processes various types of signals from various senses. The human brain
is a biological signal processing system. When the system is made up of
electronic components, it is known as an electronic signal processing
system. Signal processing is a discipline that combines mathematics and
electrical engineering.
There are two types of electronic signals—analog and digital. The
following table lists the differences between these two types of signals.

Analog Digital

A continuous signal Discrete in nature

Denoted by sinusoidal waves Denoted by square waves
A continuous signal A discrete signal
Deteriorated due to noise Not affected by noise
Not flexible Flexible and can perform a variety of operations
Consumes less bandwidth Consumes a lot of bandwidth
Draws a lot of power Draws less power compared to analog

46
 Chapter 2 Introduction to Python and Digital Image Processing

Analog Digital

There are lots of errors Fewer errors compared to analog signals,

in capturing, storing, and and there are checksum and error correction
transmitting analog signals algorithms available for digital signals.

I mage Processing
An image is a signal. So, image processing is a type of signal processing.
Image processing systems are types of signal processing systems. The
combination of the eye and brain is an example of a biological image
processing system. There are two types of image processing systems—
analog and digital.

Analog Image Processing

The days of still and motion film cameras represent the analog era. The
sources of analog images are film cameras (still and motion), older fax
machines, and telex machines. Older television systems, CRT monitors,
and film projectors represent analog image processing systems. Analog
image processing involves extensively using analog electronic systems,
mechanical parts, optics, and chemistry (to develop and store films).

Digital Image Processing

With the advent of digital computers, storage systems, image sensors, and
digital cameras, images can be captured, stored, and processed in digital
formats. Using digital computers to process and retrieve information
from an image (analog and digital images both) is known as digital image
processing. It involves extensive use of digital sensors (digital cameras),
digital computers, and digital storage devices. Here are some applications
of digital image processing systems:

47
Chapter 2 Introduction to Python and Digital Image Processing

• Computerized image editing, correction,

enhancement, denoising, etc.

• Medical image processing and diagnostics assistance

• Space image processing (processing images from

Hubble and ground-based telescopes)

• Industrial applications like product inspection

and sorting

• Biometrics (finger, face, and iris recognition)

• Filmmaking and visual effects

• Remote sensing (processing images from aerial and

satellite sources)

The following scientific disciplines significantly overlap with digital

image processing:

• Signal processing

• Digital electronics

• Computer/machine vision

• Biological vision
• Artificial intelligence, pattern recognition, and machine
learning

• Robotics and robot vision

48
 Chapter 2 Introduction to Python and Digital Image Processing

 sing Raspberry Pi and Python for Digital Image

U
Processing (DIP)
Since DIP requires digital computers, we need to use a computer and an
associated programming platform to implement digital image processing.
Raspberry Pi fulfills all the requirements of a minimal power system
required for DIP. The most appealing feature of Raspberry Pi is its low
cost. Also, it can be interfaced with a variety of digital image sensors like
webcams and the Pi camera module.
We learned that Python is an easy-to-learn programming language
that helps us focus on the task at hand rather than having to worry about
syntax. It is also the most preferred programming platform for Raspberry
Pi. Many third-party image processing and visualization libraries are
available for Python. This makes Raspberry Pi and Python the most
obvious choices for beginners to get started with DIP.

Exercise
Complete the following Exercise to better understand the Python 3
background.
• Visit and explore the Python home page at www.
python.org.
• Visit and explore the Python documentation page at
https://docs.python.org/3/.
• Check for new features of the latest releases of
Python at https://docs.python.org/3/whatsnew/
index.html.

49
Chapter 2 Introduction to Python and Digital Image Processing

Summary
In this chapter, we learned about the background, history, and features
of Python. We also studied the important differences between Python 2.x
and Python 3.x. We learned to use Python 3 in scripting and interpreter
modes. We looked at a few popular IDEs for Python and configured Geany
for Python 3 on the Pi. Finally, we learned about the field of digital image
processing. In the next chapter, we will get started with a popular digital
image processing library in Python called Pillow. We will also learn how to
use the Tkinter library to show images.

50
CHAPTER 3

Getting Started
In the previous chapter, we learned about the philosophy of Python. We
also learned why we should use Python 3, as well as several concepts
related to digital image processing. This chapter will look at image
processing programming using Python 3 on the Raspberry Pi. We will
learn how to connect a Raspberry Pi to various camera sensors in order to
acquire images. We will be introduced to the Pillow and Tkinter libraries in
Python 3. The following is a list of topics we will learn in this chapter:

• Image sources

• Using a webcam

• The Pi camera module

• Python 3 for digital image processing

By the end of this chapter, we will be comfortable using the cameras, as
well as the libraries Pillow and Tkinter.

Image Sources
To learn how to do digital image processing, we are going to need digital
images. There are standard image datasets that are used all around the
world; we will use a couple of free-to-use images from the following
database: http://sipi.usc.edu/database/.

© Ashwin Pajankar 2022 51

A. Pajankar, Raspberry Pi Image Processing Programming,
https://doi.org/10.1007/978-1-4842-8270-0_3
Chapter 3 Getting Started

We can read copyright-related information for these images at

https://sipi.usc.edu/database/copyright.php.
We are going to use image files 4.1.07.tiff, 4.1.08.tiff, gray21.512, and
ruler.512 from this database throughout the book for the demonstrations.
In this chapter, we will start programming. We need to organize all the
files in a single directory, with chapter subdirectories to save the code. Run
the following commands in terminal to create a directory structure for the
book code and image datasets:

mkdir DIP
cd DIP
mkdir code
mkdir dataset
cd code
mkdir chapter03

Extract all the images into the Dataset directory. Now the directory
structure will look like Figure 3-1.

52
 Chapter 3 Getting Started

Figure 3-1.  Directory structure for the organization of the dataset

and the code

We use the tree command in the DIP directory to view the directory
structure.

Note Please install the tree utility with the following command if

not installed already:
sudo apt install tree

The images we just downloaded and extracted are used as standard

images for all digital image processing and computer vision research
around the world.

53
Chapter 3 Getting Started

Using a Webcam
Let’s see how to capture images using a standard USB webcam. The
Raspberry Pi 3 has four USB ports. You can use one of these to connect a
webcam to the Raspberry Pi. I am using a Logitech c922 USB webcam (see
Figure 3-2).

Figure 3-2.  Logitech c9221

Note  Before purchasing a webcam, check the webcam’s

compatibility with the Pi at http://elinux.org/RPi_USB_
Webcams.

1
 Image provided by Pmwiki1 under a CC0 1.0 Universal (CC0 1.0) license
(https://creativecommons.org/publicdomain/zero/1.0/)

54
 Chapter 3 Getting Started

Attach the webcam and run the command lsusb in the terminal. This
displays a list of all USB devices connected to the computer. The output
will be similar to the following:

pi@raspberrypi:~$ lsusb
Bus 002 Device 001: ID 1d6b:0003 Linux Foundation 3.0 root hub
Bus 001 Device 003: ID 046d:085c Logitech, Inc. C922 Pro
Stream Webcam
Bus 001 Device 002: ID 2109:3431 VIA Labs, Inc. Hub
Bus 001 Device 001: ID 1d6b:0002 Linux Foundation 2.0 root hub

guvcview
If you can see the USB webcam in the output, it means that the webcam
has been detected. There are quite a few Linux utilities with which to
capture images using a webcam. If you like GUI, guvcview is one of them.
Use the following command to install it if it is not already installed:

sudo apt install guvcview

Once guvcview is installed, it can be accessed by running the following

command:

nohup guvcview &

This shows a preview window and the application window, as seen in

Figure 3-3.

55
Chapter 3 Getting Started

Figure 3-3.  The guvcview utility and camera preview window

As seen in Figure 3-3, it is a versatile tool that provides a unified

interface with which to adjust the properties of the webcam connected
to the RPi board. Images and videos can be captured with this tool. As an
exercise, please explore this tool further.

fswebcam
The other useful utility is fswebcam. It’s a command-line utility. Install it
using the following command:

sudo apt install fswebcam

You can invoke the fswebcam utility from the terminal to capture an
image with the webcam as follows:

fswebcam -r 1920x1080 --no-banner test.jpg

56
 Chapter 3 Getting Started

This will capture an image with the resolution of 1920 x 1080 pixels.
The --no-banner flag disables the timestamp banner from the image. The
image is saved as test.jpg. The output of the command is as follows:

pi@raspberrypi:~$ fswebcam -r 1920x1080 --no-banner test.jpg

--- Opening /dev/video0...
Trying source module v4l2...
/dev/video0 opened.
No input was specified, using the first.
--- Capturing frame...
Captured frame in 0.00 seconds.
--- Processing captured image...
Disabling banner.
Writing JPEG image to 'test.jpg'.

Then, check in the current directory for the image file called test.jpg.

The Pi Camera Module

The Raspberry Pi Foundation has designed dedicated camera modules for
the Raspberry Pi. There are two versions, both of which come in normal
and NoIR (No Infrared) varieties. The original versions are 5-megapixel
camera modules. The new versions are 8-megapixel camera modules.
Figure 3-4 shows an image of the camera module V2.

57
Chapter 3 Getting Started

Figure 3-4.  The Pi camera module V22

This one is the normal camera module suitable for normal lighting
conditions.
Figure 3-5 shows the NoIR version, suitable for low lighting conditions.
More details about them can be found at https://www.raspberrypi.org/
products/camera-module-v2. The camera can be connected to any model
of Raspberry Pi by a dedicated camera port, as shown in Figure 3-5.

2
 Image provided by the Raspberry Pi Foundation under a CC-by-SA 4.0 license
(https://creativecommons.org/licenses/by-sa/4.0/deed.en)

58
 Chapter 3 Getting Started

Figure 3-5.  Attaching the Pi camera module to the Pi3

In Figure 3-5, the NoIR camera module V2 is connected to a Raspberry

Pi Zero W board (which is a compact board model in the Raspberry Pi
family). The Pi camera module connects directly to the GPU. As it is
attached to the GPU, there is only a small impact on the CPU, leaving it
available for other processing. The difference between the Pi camera and
USB webcam is that the Pi camera module has a higher performance level
and higher frame rate, with h.264 video encoding.
The Raspberry Pi Foundation has produced another high-quality
camera module with interchangeable C- and CS- mounts for lenses. The

3
 Image provided by the Raspberry Pi Foundation under a CC-by-SA 4.0 license
(https://creativecommons.org/licenses/by-sa/4.0/deed.en)

59
Chapter 3 Getting Started

package includes the camera module and C-to-CS mount adapter. The
camera looks as shown in Figure 3-6.

Figure 3-6.  High-quality camera module4

We can read more about the camera module on the product page
at https://www.raspberrypi.com/products/raspberry-pi-high-
quality-camera/.

4
 Image provided by the Raspberry Pi Foundation under a CC-by-SA 4.0 license
(https://creativecommons.org/licenses/by-sa/4.0/deed.en)

60
 Chapter 3 Getting Started

raspistill
The command-line utility we use to capture images using the camera
module is raspistill. We call it as follows:

raspistill -o test.png

raspistill does not write anything to the console or terminal, so we

need to check the directory to see if test.png has been created or updated.
This is how you capture images with various cameras and command-
line utilities for use in exploring the world of DIP with Pi.

Python 3 for Digital Image Processing

Python 3 does not have any pre-installed libraries for image processing.
The Python Imaging Library (PIL) is one of the most popular beginner-
friendly image processing libraries used for image processing in Python.
This library provides extensive file-format support and reasonably
powerful image processing capabilities. The core Image library is designed
for achieving faster access to data stored in a few basic image formats. The
drawback with PIL is that after its last stable release (1.1.7) in 2009 and the
last commit in 2011, there has not been a new release. Also, PIL does not
support Python 3. Officially, we have not heard that PIL is dead. However,
there is a friendly fork (derived) project that’s an unofficial replacement and
enhancement for PIL. It works well with Python 3 and is called Pillow. It is
mostly backward compatible with PIL. You can get more information about
Pillow on its home page at https://python-pillow.org/.
Let’s get started with the basics of Pillow. We can install Pillow by
running the following command at the terminal:

sudo pip3 install pillow

61
Chapter 3 Getting Started

This will install the Pillow library for Python 3. To check if it is installed
properly, open Python 3 in interpreter mode and run the following
sequence of commands. If Pillow is installed properly, these commands
will display its version number.

Python 3.9.2 (default, Mar 12 2021, 04:06:34)

[GCC 10.2.1 20210110] on linux
Type "help", "copyright", "credits" or "license" for more
information.
>>> from PIL import Image
>>> print(Image.__version__)
8.1.2
>>>

Working with Images
Let’s work with images now. Before we begin, we will have to install an
image viewing utility called xv in order for the built-in function show()
to work. The problem is that the xv utility is deprecated. So, you will use
another utility called xli and point it to xv with a Linux command. Run the
following commands:

sudo apt-get install xli -y

cd /usr/local/bin
sudo ln -s /usr/bin/xli xv

Save the program shown in Listing 3-1 as prog01.py in the directory /

home/pi/DIP/code/chapter03.

Listing 3-1.  prog01.py

from PIL import Image

im = Image.open("/home/pi/DIP/Dataset/4.1.07.tiff")
im.show()

62
 Chapter 3 Getting Started

Run this code with the following command:

python3 prog01.py

It will show the image in an xli window. This is the simplest Pillow
program; it loads an image in a Python variable with the function call
open() and displays it with the function call show(). In the first line, we are
importing the Image module of Pillow. We will learn more about this module
in the next chapter. The standard version of show() is not very efficient,
because it saves the image to a temporary file and calls the xv utility to
display the image. However, it is handy for the purposes of debugging.
The output is as shown in Figure 3-7.

Figure 3-7.  A simple demo5

Due to the limitation of the function show(), we will use Python’s built-
in GUI module Tkinter to display images whenever needed. Listing 3-2
creates an empty window as shown in Figure 3-8.

5
 4.1.07, 4.1.08 Picture of jelly beans taken at USC. Free to use (http://sipi.usc.
edu/database/copyright.php)

63
Chapter 3 Getting Started

Listing 3-2.  prog02.py

import tkinter as tk
root = tk.Tk()
root.title("Test")
root.mainloop()

Figure 3-8.  An empty window

Let’s display an image using Tkinter, as shown in Listing 3-3.

Listing 3-3.  prog03.py

from PIL import Image, ImageTk
import tkinter as tk
im = Image.open("/home/pi/DIP/Dataset/4.1.07.tiff")
root = tk.Tk()
root.title("Test")
photo = ImageTk.PhotoImage(im)
l = tk.Label(root, image=photo)
l.pack()
l.photo = photo
root.mainloop()

64
 Chapter 3 Getting Started

This program uses Python’s built-in module for GUI, called Tkinter.
We are importing it in the second line. The ImageTk module provides the
functionality to convert the Pillow image to a Tk-compatible image with
the PhotoImage() function. We create a Tk window with the following
statement:

root = tk.Tk()

The code displays the Pillow image as a label. The method title() sets
the title of the image. In the code, the Label() and pack() functions are
used to create the image label. The line l.photo() = photo is for Python’s
garbage collector. root.mainloop() is the main loop for the Tk GUI.
The output is shown in Figure 3-9.

Figure 3-9.  Tkinter and Pillow in action

Note If you are encountering an error related to importing while

executing this program, run the following command at the terminal:
sudo apt-get install python3-pil.imagetk

65
Chapter 3 Getting Started

We can also check the properties of an image as follows:

Listing 3-4.  prog04.py

from PIL import Image

im = Image.open("/home/pi/DIP/Dataset/4.1.07.tiff")
print(im.mode)
print(im.format)
print(im.size)
print(im.info)
print(im.getbands())

It produces the following output:

RGB
TIFF
(256, 256)
{'compression': 'raw', 'dpi': (1, 1), 'resolution': (1, 1)}
('R', 'G', 'B')

Let’s look at the program line by line.

The mode of an image defines the type and depth of its pixels. Here are
the standard modes available in Pillow:
• 1 (1-bit pixels, black and white, stored with one pixel
per byte)

• L (8-bit pixels, black and white)

• P (8-bit pixels, mapped to any other mode using a color

palette)

• RGB (3x8-bit pixels, true color)

• RGBA (4x8-bit pixels, true color with

transparency mask)

66
 Chapter 3 Getting Started

• CMYK (4x8-bit pixels, color separation)

• YCbCr (3x8-bit pixels, color video format)

The format of the image refers to the file format. Size refers to the
resolution of the image in pixels. Info refers to the auxiliary information
of the image. The function getbands() retrieves the bands of the colors in
the image.

Summary
In this chapter, we got started with the basics of Pillow. We learned how
to load and display an image. We learned a bit about image properties.
We also learned how to capture images using various sensors. In the next
chapter, we will explore the modules Image, ImageChops, and ImageOps in
Pillow in detail.

67
CHAPTER 4

Basic Operations
on Images
In the previous chapter, we started with using Pillow for image processing.
We also used Tkinter for displaying images. In this chapter, we will learn
various arithmetic and logical operations to use on images. We will
explore using Image, ImageChops, and ImageOps modules in Pillow to
implement these operations. We will also learn how to use the slide bar in
Tkinter to dynamically change the program input.
We will be learning about the following topics:

• Image module

• ImageChops module
• ImageOps module

Image Module
Let’s start with the Image module. We can perform myriad operations on
images with this module.

© Ashwin Pajankar 2022 69

A. Pajankar, Raspberry Pi Image Processing Programming,
https://doi.org/10.1007/978-1-4842-8270-0_4
Chapter 4 Basic Operations on Images

Image Channels
We can use the routine split() to split an image into its constituent
channels. We can also merge various images into a single image with the
routine merge(). Listing 4-1 shows a demonstration of this.

Listing 4-1.  prog01.py

from PIL import Image, ImageTk

import tkinter as tk
im = Image.open("/home/pi/DIP/Dataset/4.1.07.tiff")
root = tk.Tk()
root.title("RED Channel Demo")
r, g, b = im.split()
photo1 = ImageTk.PhotoImage(r)
l1 = tk.Label(root, image=photo1)
l1.pack()
l1.photo = photo1
photo2 = ImageTk.PhotoImage(Image.merge("RGB", (r, g, b)))
l2 = tk.Label(root, image=photo2)
l2.pack()
l2.photo = photo2
root.mainloop()

The output is shown in Figure 4-1.

70
 Chapter 4 Basic Operations on Images

Figure 4-1.  Red channel and the original image

Colorspace Conversion
You can change the mode of an image using the routine convert(), as
shown in Listing 4-2.

Listing 4-2.  prog02.py

from PIL import Image, ImageTk
import tkinter as tk
im1 = Image.open("/home/pi/DIP/Dataset/4.1.07.tiff")
71
Chapter 4 Basic Operations on Images

res1 = im1.convert("L")
root = tk.Tk()
root.title("Colorspace Conversion Demo")
photo = ImageTk.PhotoImage(res1)
l = tk.Label(root, image=photo)
l.pack()
l.photo = photo
root.mainloop()

The code in Listing 4-2 changes the mode of the image to L. The
routine convert() supports all possible conversions between the RGB,
CMYK, and L modes. We can read more about the colorspaces at https://
www.color-­management-­guide.com/color-­spaces.html.

Image Blending
We can blend two images using the blend() method. It takes three
arguments—two images to be blended and the value of alpha. The
mathematical formula it uses for blending is as follows:
output = image1 * (1.0 - alpha) + image2 * alpha
Now, we will write a program that can change the value of alpha so
that we can experience the blending effect ourselves. We will use the scale
widget in Tkinter.
The program in Listing 4-3 demonstrates this process.

Listing 4-3.  prog03.py

from PIL import Image, ImageTk

import tkinter as tk
def show_value_1(alpha):
    print('Alpha: ', alpha)
    img = Image.blend(im1, im2, float(alpha))

72
 Chapter 4 Basic Operations on Images

    photo = ImageTk.PhotoImage(img)
    l['image'] = photo
    l.photo = photo
root = tk.Tk()
root.title('Blending Demo')
im1 = Image.open("/home/pi/DIP/Dataset/4.1.07.tiff")
im2 = Image.open("/home/pi/DIP/Dataset/4.1.08.tiff")
photo = ImageTk.PhotoImage(im1)
l = tk.Label(root, image=photo)
l.pack()
l.photo = photo
w1 = (tk.Scale(root, label="Alpha", from_=0, to=1,
      resolution=0.01, command=show_value_1, orient=tk.
HORIZONTAL))
w1.pack()
root.mainloop()

The code in Listing 4-3 creates a scale widget using the tk.Scale()
call statement. The statement has more than 79 characters, so in order to
conform to the PEP8 standard, I wrote it to fit in two lines, each consisting
of fewer than 79 characters. The parameters passed to the tk.Scale() call
are as follows:

• The Tkinter window variable name

• label: The label to be associated with scale

• from_ and to: The range of values

• resolution: The resolution of the scale bar

• command: The function to execute when the value of the

scale changes

• orient: The orientation of the scale

73
Chapter 4 Basic Operations on Images

When you change the slider with the mouse pointer, it calls the custom
show_value_1() function. We are printing the current value of the track-­
bar position to the console for debugging purposes. The statement img =
Image.blend(im1, im2, float(alpha)) creates a blended image. The
following lines update the image in the Tkinter window:

photo = ImageTk.PhotoImage(img)
l['image'] = photo
l.photo = photo

The show_value_1() function updates every time we change the slider

position and the new image is computed. This makes the program interesting
and interactive, as we can change the value of alpha to see the transition effect.
The output is shown in Figure 4-2.

Figure 4-2.  Image blending tool1

1
 4.1.07, 4.1.08 Picture of jelly beans taken at USC. Free to use (http://sipi.usc.
edu/database/copyright.php)

74
 Chapter 4 Basic Operations on Images

We will use the same basic code skeleton to demonstrate the other
features of the Pillow library.

Resizing an Image
You can resize an imageusing the routine resize(), as shown in
Listing 4-4.

Listing 4-4.  prog04.py

from PIL import Image, ImageTk

import tkinter as tk
def show_value_1(size):
    print('Resize: ', size, ' : ', size)
    img = im.resize((int(size), int(size)))
    photo = ImageTk.PhotoImage(img)
    l['image'] = photo
    l.photo = photo
root = tk.Tk()
root.attributes('-fullscreen', True)
im = Image.open("/home/pi/DIP/Dataset/4.1.07.tiff")
photo = ImageTk.PhotoImage(im)
l = tk.Label(root, image=photo)
l.pack()
l.photo = photo
w1 = (tk.Scale(root, label="Resize", from_=128,
      to=512, resolution=1, command=show_value_1, orient=tk.
HORIZONTAL))
w1.pack()
root.mainloop()

75
Chapter 4 Basic Operations on Images

This example varies the image size from (128, 128) to (512, 512). The
routine resize() takes the new size tuple as an argument. The code
also invokes the Tkinter window in full-screen mode with the root.
attributes() function call. To close this window, you have to press Alt+F4
from the keyboard. Run the code and have a look at the output.

Rotating an Image
We can use the routine rotate(), which takes the angle of rotation as an
argument. The code in Listing 4-5 demonstrates this idea.

Listing 4-5.  prog05.py

from PIL import Image, ImageTk

import tkinter as tk
def show_value_1(angle):
    print('Angle: ', angle)
    img = im.rotate(float(angle))
    photo = ImageTk.PhotoImage(img)
    l['image'] = photo
    l.photo = photo
root = tk.Tk()
root.title("Rotation Demo")
im = Image.open("/home/pi/DIP/Dataset/4.1.07.tiff")
photo = ImageTk.PhotoImage(im)
l = tk.Label(root, image=photo)
l.pack()
l.photo = photo
w1 = (tk.Scale(root, label="Angle", from_=0, to=90,
      resolution=1, command=show_value_1, orient=tk.
HORIZONTAL))
w1.pack()
root.mainloop()

76
 Chapter 4 Basic Operations on Images

Run this code to experience the rotation effect on an image. We can

also transpose the images using the transpose() function. It takes one of
PIL.Image.FLIP_LEFT_RIGHT, PIL.Image.FLIP_TOP_BOTTOM, PIL.Image.
ROTATE_90, PIL.Image.ROTATE_180, PIL.Image.ROTATE_270, or PIL.
Image.TRANSPOSE as an argument. The code example in Listing 4-6 shows
a rotation at 180 degrees.

Listing 4-6.  prog06.py

from PIL import Image, ImageTk

import tkinter as tk
root = tk.Tk()
root.title("Transpose Demo")
im = Image.open("/home/pi/DIP/Dataset/4.1.07.tiff")
out = im.transpose(Image.ROTATE_180)
photo = ImageTk.PhotoImage(out)
l = tk.Label(root, image=photo)
l.pack()
l.photo = photo
root.mainloop()

Also, the following line (prog07.py in the code bundle) will flip the
image vertically:

out = im.transpose(Image.FLIP_TOP_BOTTOM)

Crop and Paste Operations

We can crop a part of an image using the crop() method. It takes an
argument of four tuples that specify the coordinates of the box to be
cropped from the image. Pillow also has a paste() method to paste a
rectangular image to another image. The paste() method takes the image

77
Chapter 4 Basic Operations on Images

to be pasted and the coordinates as arguments. The program in Listing 4-7

demonstrates how we can rotate a face using a clever combination of
crop(), rotate(), and paste().

Listing 4-7.  prog08.py

from PIL import Image

im = Image.open("/home/pi/DIP/Dataset/4.1.07.tiff")
face_box = (100, 100, 150, 150)
face = im.crop(face_box)
rotated_face = face.transpose(Image.ROTATE_180)
im.paste(rotated_face, face_box)
im.show()

The output of the code in Listing 4-7 is shown in Figure 4-3.

Figure 4-3.  Crop and paste demo

78
 Chapter 4 Basic Operations on Images

Copying and Saving Images to a File

You can use the copy() method to copy an entire Pillow image to another
Python variable. You can save a Pillow image to a file using the save()
method. A demonstration is shown in Listing 4-8.

Listing 4-8.  prog09.py

from PIL import Image

im = Image.open("/home/pi/DIP/Dataset/4.1.07.tiff")
im1 = im.copy()
im1.save("test.tiff")

The code in Listing 4-8 opens an image from a given location, copies it
into the im1 variable, and saves it to the current location as test.tiff.

Knowing the Value of a Particular Pixel

You can determine the value of a particular pixel using getpixel(). It is
usually a tuple that represents the channel intensities. With RGB images,
you get the Red, Green, and Blue intensities. It is used (prog10.py in the
code bundle) as follows:

print(im.getpixel((100,100)))

This way, we can determine the value of the pixel.

Mandelbrot Set
A Mandelbrot set is a fractal, and we can generate it with Pillow.

Note Read more about Mandelbrot set at ­https://mathworld.

wolfram.com/MandelbrotSet.html

79
Chapter 4 Basic Operations on Images

We can create a Mandelbrot set with the code in Listing 4-9.

Listing 4-9.  prog10_1.py

from PIL import Image

# drawing area (xa < xb and ya < yb)
xa = -2.0
xb = 1.0
ya = -1.5
yb = 1.5
maxIt = 256 # iterations
# image size
imgx = 512
imgy = 512
#create mtx for optimized access
image = Image.new("RGB", (imgx, imgy))
mtx = image.load()
#optimizations
lutx = [j * (xb-xa) / (imgx - 1) + xa for j in range(imgx)]
for y in range(imgy):
    cy = y * (yb - ya) / (imgy - 1)  + ya
    for x in range(imgx):
        c = complex(lutx[x], cy)
        z = 0
        for i in range(maxIt):
            if abs(z) > 2.0: break
            z = z * z + c
        r = i % 4 * 64
        g = i % 8 * 32
        b = i % 16 * 16
        mtx[x, y] =  r,g,b
image.show()

80
 Chapter 4 Basic Operations on Images

image.save("mandel.png", "PNG")

The output is shown in Figure 4-4.

Figure 4-4.  Mandelbrot set

Julia Set
We can create a Julia set as demonstrated in Listing 4-10.

81
Chapter 4 Basic Operations on Images

Listing 4-10.  prog10_2.py

from PIL import Image

import random
# drawing area (xa < xb and ya < yb)
xa = -2.0
xb = 1.0
ya = -1.5
yb = 1.5
maxIt = 256 # iterations
# image size
imgx = 512
imgy = 512
image = Image.new("RGB", (imgx, imgy))
# Julia set to draw
c = complex(random.random() * 2.0 - 1.0, random.random() - 0.5)
for y in range(imgy):
    zy = y * (yb - ya) / (imgy - 1)  + ya
    for x in range(imgx):
        zx = x * (xb - xa) / (imgx - 1) + xa
        z = complex(zx, zy)
        for i in range(maxIt):
            if abs(z) > 2.0: break
            z = z * z + c
        r = i % 4 * 64
        g = i % 8 * 32
        b = i % 16 * 16
        image.putpixel((x, y), b * 65536 + g * 256 + r)
image.show()
image.save("julia.png", "PNG")

The output is as shown in Figure 4-5.

82
 Chapter 4 Basic Operations on Images

Figure 4-5.  Julia set

Note  You can read more about Julia sets at the following URLs:
https://mathworld.wolfram.com/JuliaSet.html
https://www.britannica.com/science/Julia-­set

83
Chapter 4 Basic Operations on Images

Noise and Gradients
Noise is the unwanted part in a signal. Depending on the type of signal,
noise can take various forms. We know that an image is a signal. So, an
image can have noise. Noise can be inadvertently added into an image
at the time of capturing that image or transmission. We can generate (or
simulate) noise as shown in Listing 4-11.

Listing 4-11.  prog10_3.py

from PIL import Image

im = Image.effect_noise((512, 512), 0.01)
im.show()

We can generate a linear gradient as demonstrated in Listing 4-12.

Listing 4-12.  prog10_4.py

from PIL import Image

im = Image.linear_gradient('P')
im.show()

We can generate a radial gradient as demonstrated in Listing 4-13.

Listing 4-13.  prog10_5.py

from PIL import Image

im = Image.radial_gradient('P')
im.show()

Please run all these listings to see the code in action.

84
 Chapter 4 Basic Operations on Images

ImageChops Module
This module has routines for many basic arithmetic and logical operations
that we can use on images. Let’s look at them quickly one by one.
We can add two images together using the add() method.
The following is sample code for adding images:

im3 = ImageChops.add(im1, im2)

The add_module() method adds two images without clipping

the result:

im3 = ImageChops.add_modulo(im1, im2)

The darker() method compares two images, pixel by pixel, and

returns the darker pixels:

im3 = ImageChops.darker(im1, im2)

The difference() method returns the difference of the absolute

values of two images:

im3 = ImageChops.difference(im1, im2)

It uses the following mathematical formula for calculating the

difference:
image3 = abs(image1 - image2)
We can invert an image as follows:

im2 = ImageChops.invert(im1)

Just like with darker(), you can use the lighter() method to return
the set of lighter pixels:

im3 = ImageChops.lighter(im1, im2)

85
Chapter 4 Basic Operations on Images

logical_and() and logical_or() are the logical operations to use on

images. These are explained with the help of black-and-white images. The
following are the example usages of these routines:

im1 = Image.open("/home/pi/DIP/Dataset/ruler.512.tiff")
im2 = Image.open("/home/pi/DIP/Dataset/ruler.512.tiff")
im2 = im2.transpose(Image.ROTATE_90)
im3 = ImageChops.logical_and(im1.convert("1"), im2.
convert("1"))
im3 = ImageChops.logical_or(im1.convert("1"), im2.convert("1"))

These examples convert the grayscale images to black-and-white

images first and then perform the logical operations on them. Figure 4-6
shows the output of logical operation AND.

86
 Chapter 4 Basic Operations on Images

Figure 4-6.  Output of logical operations on black-and-white images2

You can superimpose one image on another using the

multiply() method:

im3 = ImageChops.multiply(im1, im2)

2
 gray21.512, ruler.512 Test patterns constructed at USC-­SIPI. Free to use.
(http://sipi.usc.edu/database/copyright.php)

87
Chapter 4 Basic Operations on Images

The screen() method superimposes inverted images on top of

each other:

im3 = ImageChops.screen(im1, im2)

You can subtract one image from another using the subtract()
method as follows:

im3 = ImageChops.subtract(im1, im2)

You can subtract without clipping the result as follows:

im3 = ImageChops.subtract_modulo(im1, im2)

You can fill a channel with a given gray level as follows:

im2 = ImageChops.constant(im1, 10)

You can compute a logical exclusive OR as follows:

im3 = ImageChops.logical_xor(im1.convert("1"),
                             im2.convert("1"))

You can superimpose two images on top of each other using the Soft
Light algorithm as follows:

im3 = ImageChops.soft_light(im1, im2)

You can superimpose two images on top of each other using the Hard
Light algorithm as follows:

im3 = ImageChops.hard_light(im1, im2)

You can superimpose two images on top of each other using the
Overlay algorithm as follows:

im3 = ImageChops.overlay(im1, im2)

You can also offset the data of an image as follows:

88
 Chapter 4 Basic Operations on Images

im2 = ImageChops.offset(im1,
                        xoffset=128,
                        yoffset=128)

ImageOps
This module has many predefined and useful operations.
You can automatically adjust the contrast of an image as follows:

im2 = ImageOps.autocontrast(im1)

You can colorize a grayscale image as follows:

im2 = ImageOps.colorize(im1, (125, 150, 134), (10, 12, 17))

You can crop the borders of an image equally from all sides as follows:

im2 = ImageOps.crop(im1, 50)

The first argument of the ImageOps.crop() method is the image, and

the second argument is the width of the cropped border in pixels.
You can also expand the border of an image. Expanded borders will be
filled with black pixels equally on all sides, as follows:

im2 = ImageOps.expand(im1, 50)

You can flip an image vertically as follows:

im2 = ImageOps.flip(im1)

You can also flip it horizontally as follows:

im2 = ImageOps.mirror(im1)

You can reduce the number of bits of all the color channels using the
posterize() method. This takes the image and the number of bits to keep
for every channel as arguments. The following is an example:

89
Chapter 4 Basic Operations on Images

im2 = ImageOps.posterize(im1, 3)

This example keeps only three bits per channel.

The solarize() method inverts all the pixels above a particular
grayscale threshold as follows:

im2 = ImageOps.solarize(im1, 100)

You can pad an image as follows:

im2 = ImageOps.pad(im1, (128, 128))

You can scale an image as follows:

im2 = ImageOps.scale(im1, 1.5)

You can automatically equalize the image histogram as follows:

im2 = ImageOps.equalize(im1)

You can convert a color image to a grayscale image as follows:

im2 = ImageOps.grayscale(im1)

You can invert an image as follows:

im2 = ImageOps.invert(im1)

Summary
This chapter explored the modules Image, ImageChops, and ImageOps
in detail. In the next chapter, we will explore a few more Pillow modules
for advanced operations on images, such as filtering, enhancements,
histograms, and quantization.

90
CHAPTER 5

Advanced Operations
on Images
The previous chapter explored the arithmetic and logical operations to use
on images with Pillow. There is more to the world of image processing than
that, however. Pillow has a lot more functionality to offer. You can enhance
and filter images. You can also calculate the histogram of an image as well
as its channels. We will cover the following topics in this chapter:

• ImageFilter module

• ImageEnhance module

• Color quantization

• Histograms and equalization

ImageFilter Module
We can use the ImageFilter module in Pillow to perform a variety of
filtering operations on images. You can use filters to remove noise, to add
blur effects, and to sharpen and smooth your images. Listing 5-1 shows a
simple image-filtering program that uses the ImageFilter module.

© Ashwin Pajankar 2022 91

A. Pajankar, Raspberry Pi Image Processing Programming,
https://doi.org/10.1007/978-1-4842-8270-0_5
Chapter 5 Advanced Operations on Images

Listing 5-1.  prog01.py

from PIL import Image, ImageFilter

im1 = Image.open("/home/pi/DIP/Dataset/4.1.08.tiff")
im2 = im1.filter(ImageFilter.BLUR)
#im2 = im1.filter(ImageFilter.CONTOUR)
#im2 = im1.filter(ImageFilter.DETAIL)
#im2 = im1.filter(ImageFilter.EDGE_ENHANCE)
#im2 = im1.filter(ImageFilter.EDGE_ENHANCE_MORE)
#im2 = im1.filter(ImageFilter.EMBOSS)
#im2 = im1.filter(ImageFilter.FIND_EDGES)
#im2 = im1.filter(ImageFilter.SMOOTH)
#im2 = im1.filter(ImageFilter.SMOOTH_MORE)
#im2 = im1.filter(ImageFilter.SHARPEN)
im2.show()

Figure 5-1 shows the output of this program.

Figure 5-1.  Blur filter demo

Uncomment the lines in the code file and see the other filters in action.

92
 Chapter 5 Advanced Operations on Images

We can also have a custom filter, as shown in Listing 5-2.

Listing 5-2.  prog02.py

from PIL import Image, ImageFilter

im1 = Image.open("/home/pi/DIP/Dataset/4.1.08.tiff")
custom_filter = ImageFilter.GaussianBlur(radius=float(5))
im2 = im1.filter(custom_filter)
im2.show()

Figure 5-2 shows the output of this program.

Figure 5-2.  Custom Gaussian filter

The code in Listing 5-2 uses the GaussianBlur() method for a custom
filter. The radius of the blur is 5. We can change the radius. Let’s use a
slider in Tkinter to modify the code and make the blur radius dynamic
(see Listing 5-3).

93
Chapter 5 Advanced Operations on Images

Listing 5-3.  prog03.py

from PIL import Image, ImageTk, ImageFilter

import tkinter as tk
def show_value_1(blur_radius):
    print('Gaussian Blur Radius: ', blur_radius)
    custom_filter = ImageFilter.GaussianBlur(radius=float(blur_
radius))
    img = im1.filter(custom_filter)
    photo = ImageTk.PhotoImage(img)
    l['image'] = photo
    l.photo = photo
root = tk.Tk()
root.title('Gaussian Blur Filter Demo')
im1 = Image.open("/home/pi/DIP/Dataset/4.1.07.tiff")
photo = ImageTk.PhotoImage(im1)
l = tk.Label(root, image=photo)
l.pack()
l.photo = photo
w1 = (tk.Scale(root, label="Blur Radius", from_=0, to=10,
      resolution=0.2, command=show_value_1, orient=tk.
HORIZONTAL))
w1.pack()
root.mainloop()

The output of Listing 5-3 is shown in Figure 5-3.

94
 Chapter 5 Advanced Operations on Images

Figure 5-3.  Gaussian blur with TKinter

We can see the GaussianBlur() has been applied on the image.

The next filter we will explore is the convolution filter. We need to
understand the concept of kernels for this example. Kernels are the square
matrices used in image processing operations. We mostly use kernels
for filtering. Kernels produce a variety of effects, including blurring,
smoothing, noise reduction, sharpening, and edge highlighting.
When filters are used to remove high-frequency components from
an image, the process is called low-pass filtering, since it allows the low-­
frequency data to pass through the filter. It creates effects like blurring,
smoothing, and noise reduction.
When filters are used to remove low-frequency components from an
image, the process is called high-pass filtering, since it allows the high-­
frequency data to pass through the filter. It creates effects like sharpening
and edge highlighting.
Figure 5-4 shows an example kernel.

95
Chapter 5 Advanced Operations on Images

Figure 5-4.  A kernel

Listing 5-4 shows a code example of the kernel used for image
convolution.

Listing 5-4.  prog04.py

from PIL import Image, ImageTk, ImageFilter
import tkinter as tk
root = tk.Tk()
root.title('Convolution Kernel Demo')
im1 = Image.open("/home/pi/DIP/Dataset/4.1.07.tiff")
custom_filter = ImageFilter.Kernel((3, 3), [1, 1, 1, 1, -4, 1,
1, 1, 1])
img = im1.filter(custom_filter)
photo = ImageTk.PhotoImage(img)
l = tk.Label(root, image=photo)
l.pack()
l.photo = photo
root.mainloop()

96
 Chapter 5 Advanced Operations on Images

Run the program in Listing 5-4. It produces the following

(Figure 5-5) output.

Figure 5-5.  A convolution demo

As of now, the ImageFilter.Kernel() method only supports the

kernels of sizes 3x3 and 5x5. Try the following 5x5 kernel:

custom_filter = ImageFilter.Kernel((5, 5), [1,1,1,1,1,

1,1,1,1,1, 1,1,-10,1,1, 1,1,1,1,1, 1,1,1,1,1])

We can use the Digital Unsharp Mask filter, as shown in Listing 5-5.

Listing 5-5.  prog05.py

from PIL import Image, ImageTk, ImageFilter

import tkinter as tk
def show_value_1(blur_radius):
    print('Unsharp Blur Radius: ', blur_radius)
    custom_filter = ImageFilter.UnsharpMask(radius=float(blur_
radius))
    img = im1.filter(custom_filter)

97
Chapter 5 Advanced Operations on Images

    photo = ImageTk.PhotoImage(img)
    l['image'] = photo
    l.photo = photo
root = tk.Tk()
root.title('Digital Unsharp Mask Demo')
im1 = Image.open("/home/pi/DIP/Dataset/4.1.07.tiff")
photo = ImageTk.PhotoImage(im1)
l = tk.Label(root, image=photo)
l.pack()
l.photo = photo
w1 = (tk.Scale(root, label="Blur Radius", from_=0, to=10,
      resolution=0.2, command=show_value_1, orient=tk.
HORIZONTAL))
w1.pack()
root.mainloop()

The digital unsharpening mask sharpens the image. Figure 5-6 shows

the output with a mask radius of size 10.

98
 Chapter 5 Advanced Operations on Images

Figure 5-6.  Digital unsharp demo

The radius of size 10 creates a highly sharpened image.

Let’s study the median, min, and max filters in Pillow. All these filters
only accept an odd number as the window size. The code in Listing 5-6
demonstrates a median filter.

Listing 5-6.  prog06.py

from PIL import Image, ImageTk, ImageFilter

import tkinter as tk
def show_value_1(window_size):
    print('Window Size: ', window_size)
    if (int(window_size) % 2 == 0):
        print("Invalid Window Size...\nWindow Size must be an
odd number...")
    else:

99
Chapter 5 Advanced Operations on Images

        custom_filter = ImageFilter.
MedianFilter(size=int(window_size))
        img = im1.filter(custom_filter)
        photo = ImageTk.PhotoImage (img)
        l['image'] = photo
        l.photo = photo
root = tk.Tk()
root.title('Median Filter Demo')
im1 = Image.open("/home/pi/DIP/Dataset/4.1.07.tiff")
photo = ImageTk.PhotoImage(im1)
l = tk.Label(root, image=photo)
l.pack()
l.photo = photo
w1 = (tk.Scale(root, label="Window Size", from_=1, to=19,
      resolution=1, command=show_value_1, orient=tk.
HORIZONTAL))
w1.pack()
root.mainloop()

The output shown in Figure 5-7 has a window size of 19.

100
 Chapter 5 Advanced Operations on Images

Figure 5-7.  A median filter

If you change the filter to a min filter using the following code line

custom_filter = ImageFilter.MinFilter(size=int(window_size))

you’ll get the output shown in Figure 5-8.

101
Chapter 5 Advanced Operations on Images

Figure 5-8.  A min filter

Here’s an example of the max filter:

custom_filter = ImageFilter.MaxFilter(size=int(window_size))

The output for the max filter is shown in Figure 5-9.

102
 Chapter 5 Advanced Operations on Images

Figure 5-9.  The max filter

Next, we will see an example of a mode filter. The mode filter works
with even and odd window sizes. Listing 5-7 shows a code example of the
mode filter.

Listing 5-7.  prog09.py

from PIL import Image, ImageTk, ImageFilter
import tkinter as tk
def show_value_1(window_size):
    print('Window Size: ', window_size)
    custom_filter = ImageFilter.
ModeFilter(size=int(window_size))
    img = im1.filter(custom_filter)
    photo = ImageTk.PhotoImage(img)
    l['image'] = photo
    l.photo = photo

103
Chapter 5 Advanced Operations on Images

root = tk.Tk()
root.title('Mode Filter Demo')
im1 = Image.open("/home/pi/DIP/Dataset/4.1.07.tiff")
photo = ImageTk.PhotoImage(im1)
l = tk.Label(root, image=photo)
l.pack()
l.photo = photo
w1 = (tk.Scale(root, label="Window Size", from_=1, to=19,
      resolution=1, command=show_value_1, orient=tk.
HORIZONTAL))
w1.pack()
root.mainloop()

The output is shown in Figure 5-10.

Figure 5-10.  The mode filter

104
 Chapter 5 Advanced Operations on Images

We can also use the code for BoxBlur() as follows:

custom_filter = ImageFilter.BoxBlur(radius=int(window_size))

The output is shown in Figure 5-11.

Figure 5-11.  The box blur

You’ve seen how to utilize filters; now, let’s take a look at how images
can be enhanced.

The ImageEnhance Module

We can use the ImageEnhance module in Pillow to adjust the contrast,
color, sharpness, and brightness of an image, just like we used to do in old
analog television sets.
Listing 5-8 shows the code for color adjustment.

105
Chapter 5 Advanced Operations on Images

Listing 5-8.  prog11.py

from PIL import Image, ImageTk, ImageEnhance

import tkinter as tk
def show_value_1(factor):
    print('Color Factor: ', factor)
    enhancer = ImageEnhance.Color(im1)
    img = enhancer.enhance(float(factor))
    photo = ImageTk.PhotoImage(img)
    l['image'] = photo
    l.photo = photo
root = tk.Tk()
root.title('Color Adjustment Demo')
im1 = Image.open("/home/pi/DIP/Dataset/4.1.07.tiff")
photo = ImageTk.PhotoImage(im1)
l = tk.Label(root, image=photo)
l.pack()
l.photo = photo
w1 = (tk.Scale(root, label="Color Factor", from_=0, to=2,
      resolution=0.1, command=show_value_1, orient=tk.
HORIZONTAL))
w1.pack()
w1.set(1)
root.mainloop()

In Listing 5-8, the image processing code is as follows:

enhancer = ImageEnhance.Color(im1)
img = enhancer.enhance(float(factor))

We can follow the same style of coding for all the other image
enhancement operations. First, we create an enhancer, and then we apply
the enhancement factor to that. We also must have observed w1.set(1) in

106
 Chapter 5 Advanced Operations on Images

the code. This sets the scale to 1 at the beginning. Changing the argument
to set() changes the default position of the scale pointer.
Run the program in Listing 5-8 and take a look at the output.
To change the contrast, use the code in Listing 5-9.

Listing 5-9.  prog12.py

from PIL import Image, ImageTk, ImageEnhance

import tkinter as tk
def show_value_1(factor):
    print('Contrast Factor: ', factor)
    enhancer = ImageEnhance.Contrast(im1)
    img = enhancer.enhance(float(factor))
    photo = ImageTk.PhotoImage(img)
    l['image'] = photo
    l.photo = photo
root = tk.Tk()
root.title('Contrast Adjustment Demo')
im1 = Image.open("/home/pi/DIP/Dataset/4.1.07.tiff")
photo = ImageTk.PhotoImage(im1)
l = tk.Label(root, image=photo)
l.pack()
l.photo = photo
w1 = (tk.Scale(root, label="Contrast Factor", from_=0, to=2,
      resolution=0.1, command=show_value_1, orient=tk.
HORIZONTAL))
w1.pack()
w1.set(1)
root.mainloop()

Run the program in Listing 5-9 and take a look at the output.
The following enhancer is used to change the brightness:

107
Chapter 5 Advanced Operations on Images

enhancer = ImageEnhance.Brightness(im1)

The following enhancer is used to change the sharpness:

enhancer = ImageEnhance.Sharpness(im1)

For finer control of the sharpness, use the following code for the scale:

w1 = (tk.Scale(root, label="Sharpness Factor", from_=0, to=2,

      resolution=0.1, command=show_value_1, orient=tk.
HORIZONTAL))

These were all image enhancement operations. The next section looks
at more advanced image operations.

Color Quantization
Color quantization is the process of reducing the number of distinct colors
in an image. The new image should be similar to the original image in
appearance. Color quantization is done for a variety of purposes, including
when you want to store an image in a digital medium. Real-life images
have millions of colors. However, encoding them in the digital format and
retaining all the color-related information would result in a huge image
size. If you limit the number of colors in the image, you’ll need less space
to store the color-related information. This is the practical application of
quantization. The Image module has a method called quantize() that’s
used for image quantization.
The code in Listing 5-10 demonstrates image quantization in Pillow.

Listing 5-10.  prog15.py

from PIL import Image, ImageTk

import tkinter as tk
def show_value_1(num_of_col):

108
 Chapter 5 Advanced Operations on Images

    print('Number of colors: ', num_of_col)

    img = im1.quantize(int(num_of_col))
    photo = ImageTk.PhotoImage(img)
    l['image'] = photo
    l.photo = photo
root = tk.Tk()
root.title('Color Quantization Demo')
im1 = Image.open("/home/pi/DIP/Dataset/4.1.07.tiff")
photo = ImageTk.PhotoImage(im1)
l = tk.Label(root, image=photo)
l.pack()
l.photo = photo
w1 = (tk.Scale(root, label="Number of colors", from_=4, to=16,
      resolution=1, command=show_value_1, orient=tk.
HORIZONTAL))
w1.pack()
w1.set(256)
root.mainloop()

The following output (Figure 5-12) shows a quantized image.

109
Chapter 5 Advanced Operations on Images

Figure 5-12.  A quantized image

Histograms and Equalization
You likely studied frequency tables in statistics in school. Well, the
histogram is nothing but a frequency table visualized. You can calculate
the histogram of any dataset represented in the form of numbers.
The Image module has a method called histogram() that’s used
to calculate the histogram of an image. An RGB image has three 8-bit
channels. This means that it can have a staggering 256 x 256 x 256 number
of colors. Drawing a histogram of such a dataset would be very difficult. So,
the histogram() method calculates the histogram of individual channels
in an image. Each channel has 256 distinct intensities. The histogram is a
list of values for each intensity level of a channel.
The histogram for each channel has 256 numbers in the list. Suppose
the histogram for the Red channel has the values (4, 8, 0, 19, …, 90). This

110
 Chapter 5 Advanced Operations on Images

means that four pixels have the red intensity of 0, eight pixels have the red
intensity of 1, no pixel has red intensity of 2, 19 pixels have the red intensity
of three, and so on, until the last value, which indicates that 90 pixels have
the red intensity of 255.
When we combine the histogram of all three channels, we get a list of
768 numbers. In this chapter, we will just compute the histogram. We will
not show it visually. When we learn about the advanced image processing
library scipy.ndimage, we will learn how to represent histograms for each
channel individually.
The code in Listing 5-11 calculates and stores the histograms of an
image and its individual channels.

Listing 5-11.  prog16.py

from PIL import Image

im1 = Image.open("/home/pi/DIP/Dataset/4.1.07.tiff")
print(len(im1.histogram()))
r, g, b = im1.split()
print(len(r.histogram()))
print(len(g.histogram()))
print(len(b.histogram()))

Modify this program to directly print the histograms of the image and
channels.
A grayscale image (L mode image) will have a histogram of only 256
values because it has only a single channel.

Histogram Equalization
You can adjust the histogram to enhance the image contrast. This
technique is known as histogram equalization. The ImageOps.equalize()
method equalizes the histogram of the image. Listing 5-12 shows an
example of this process.

111
Chapter 5 Advanced Operations on Images

Listing 5-12.  prog17.py

from PIL import Image, ImageOps

im1 = Image.open("/home/pi/DIP/Dataset/4.1.07.tiff")
print(im1.histogram())
im2 = ImageOps.equalize(im1)
print(im2.histogram())
im2.show()

The program in Listing 5-12 prints the histogram of the original image
after the equalization. Add the im1.show() statement to the program and
then run it to see the difference between the images.

Summary
In this chapter, we explored how to use the Pillow library for advanced
image processing. Pillow is good for the beginners who want to get started
with an easy-to-program and less mathematical image processing library.
However, if you want a more mathematical and scientific approach, then
Pillow might not be your best choice. In the following chapters, we will
learn about a more powerful library to use for image processing, scipy.
ndimage. It is widely used by the scientific community all over the world.
We will also learn the basics of the NumPy and matplotlib libraries, which
come in handy when processing and displaying images.

112
CHAPTER 6

Introduction to the
Scientific Python
Ecosystem
In the previous chapter, we studied advanced image processing with
Pillow. Pillow is a nice starting point for image processing operations.
However, it has its limitations. When it comes to implementing elaborate
image processing operations, such as segmentation, morphological
operations, advanced filters, and measurements, Pillow proves to be
inadequate. We really need to use better libraries for advanced image
processing. The SciPy ecosystem serves as a foundation for all the
scientific uses of Python.
SciPy stands for Scientific Python. It extensively uses NumPy
(Numerical Python) and matplotlib for numeric operations and data
visualization, respectively. This chapter covers the basics of NumPy, SciPy,
and matplotlib. It also explores introductory-level programming examples
using NumPy, scipy.misc, and matplotlib. We will explore the following
topics in this chapter:

© Ashwin Pajankar 2022 113

A. Pajankar, Raspberry Pi Image Processing Programming,
https://doi.org/10.1007/978-1-4842-8270-0_6
Chapter 6 Introduction to the Scientific Python Ecosystem

• The Scientific Python ecosystem

• matplotlib

• Conversion between PIL image objects and NumPy

ndarrays

By the end of this chapter, we will be comfortable with the basics of the
SciPy ecosystem.

The Scientific Python Ecosystem

The Scientific Python ecosystem is an open source and free collection of
libraries for Python to be used in the scientific domain.
It has modules for numerical operations (NumPy), scientific
operations (SciPy), and data visualization (matplotlib). It is used
extensively by scientific, mathematical, engineering, and analytics
communities around the world.
The ecosystem has the following core components:

• NumPy

• SciPy
• matplotlib

• IPython

• Pandas

• Sympy

• Nose

Many other libraries use the core modules in the ecosystem for
additional algorithms and data structures. Examples include OpenCV
and Scikit.

114
 Chapter 6 Introduction to the Scientific Python Ecosystem

Figure 6-1 aptly summarizes the role of the Scientific Python ecosystem
in the world of scientific computing.

Figure 6-1.  Core components of the Scientific Python ecosystem

The best way to install the components of the Scientific Python

ecosystem on a Raspberry Pi is to use the pip utility.
First, upgrade the Pi with the following command:

sudo python3 -m pip install --upgrade pip

Then, install the components of the Scientific Python ecosystem with

the following command:

pip3 install numpy scipy matplotlib ipython jupyter pandas

sympy nose

This installs all the required libraries.

Simple Examples
Let’s study a few simple examples. Let’s learn how to read a built in image
from scipy.misc, store it in a variable, and see the data type of the variable
as shown in Listing 6-1,

115
Chapter 6 Introduction to the Scientific Python Ecosystem

Listing 6-1.  prog01.py

from scipy import misc

img = misc.face()
print(type(img))

The output is as follows:

<class 'numpy.ndarray'>

All the variables that store image data belong to this type as far as
Scientific Python is concerned. This means that the data type of the image
is ndarray in NumPy. To get started with scientific image processing,
and any type of scientific programming in general, we must know what
NumPy is.
The NumPy homepage at www.numpy.org says this:
NumPy is the fundamental package for scientific computing
with Python.
It offers the following features:
• A powerful multi-dimensional array object

• Useful methods for mathematical computations

• Wrappers and tools for integration with faster C/C++
and FORTRAN code

To get started with image processing using SciPy and NumPy, we need
to learn the basics of N-dimensional (or multi-dimensional) array objects
in NumPy.
NumPy’s N-dimensional array is a homogeneous (contains elements
all of the same data type) multi-dimensional array. It has multiple
dimensions. Each dimension is known as an axis. The class corresponding
to the N-dimensional array in NumPy is numpy.ndarray. This is what
we saw in the output of the Listing 6-1. All the major image processing

116
 Chapter 6 Introduction to the Scientific Python Ecosystem

and computer vision libraries, like Mahotas, OpenCV, scikit-image,

and scipy.ndimage (we will study this last one extensively in this book),
use numpy.ndarray to represent images. All these libraries have read(),
open(), and imread() methods for loading images from disk to a numpy.
ndarray object.
NumPy and N-dimensional arrays in NumPy are such vast topics
themselves that it would require volumes of books to explain them fully.
Hence, we will learn the relevant and important features of these as and
when they’re needed. For now, you need to understand a few important
ndarray properties that will help you comprehend important attributes of
the images that ndarray represents.
We can also check the properties of an image. Consider the code in
Listing 6-2.

Listing 6-2.  prog02.py

from scipy import misc

img = misc.face()
print(img.dtype)
print(img.shape)
print(img.ndim)
print(img.size)

The output is as follows:

uint8
(768, 1024, 3)
3
2359296

Let’s look at what each of these means. The dtype attribute is for the
data type of the elements that represent the image. In this case, it is uint8,
which means an unsigned 8-bit integer. This means it can have 256 distinct
values. shape means the dimension or size of the images. In this case, it is

117
Chapter 6 Introduction to the Scientific Python Ecosystem

a color image. Its resolution is 1024 x 768, and it has three color channels
corresponding to the colors red, green, and blue. Each channel for each
pixel can have one of the 256 possible values. So, a combination can
produce 256*256*256 distinct colors for each pixel.
You can visualize a color image as an arrangement of three two-­
dimensional planes. A grayscale image is a single plane of grayscale values.
ndim represents the dimensions. A color image has three dimensions, and
a grayscale image has two dimensions. size represents the total number of
elements in the array. It can be calculated by multiplying the values of the
dimensions. In this case, it is 768*1024*3 = 2359296.
We can see the RGB value corresponding to each individual pixel, as
shown in Listing 6-3.

Listing 6-3.  prog03.py

from scipy import misc

img = misc.face()
print(img[10, 10])

The code in Listing 6-3 accesses the value of the pixel located at (10, 10).
The output is [172 169 188].
This concludes the basics of NumPy and image processing. You will
learn more about NumPy as and when needed throughout the chapters.

Matplotlib
We have thus far used the misc.imshow() method to display an image.
While this method is useful for simple applications, it is primitive. We must
use a more advanced framework for scientific applications. Matplotlib
serves this purpose. It is a Matlab-style plotting and data visualization
library for Python. We have already installed it. It is an integral part of the
Scientific Python ecosystem. Just like NumPy, matplotlib is a vast topic

118
 Chapter 6 Introduction to the Scientific Python Ecosystem

and warrants a dedicated book. The examples in this book use the pyplot
module in matplotlib for the image processing requirements. Listing 6-4
shows a simple program for image processing.

Listing 6-4.  prog04.py

import scipy.misc as misc

import matplotlib.pyplot as plt
img = misc.face()
plt.imshow(img)
plt.show()

The code in Listing 6-4 imports the pyplot module. The imshow()
method adds the image to the plot window. The show() method shows the
plot window. The output is shown in Figure 6-2.

119
Chapter 6 Introduction to the Scientific Python Ecosystem

Figure 6-2.  Matplotlib demo1

We can also turn off the axes (or the ruler) and add a title to the image,
as shown in Listing 6-5.

Listing 6-5.  prog05.py

import scipy.misc as misc

import matplotlib.pyplot as plt
img = misc.ascent()

1
 Image provided by Judy Weggelaar (This file is in public domain, not copyrighted,
no rights reserved, free for any use.) License: https://commons.wikimedia.org/
wiki/File:Raccoon_procyon_lotor.jpg

120
 Chapter 6 Introduction to the Scientific Python Ecosystem

plt.imshow(img, cmap='gray')
plt.axis('off')
plt.title('Ascent')
plt.show()

As the image is grayscale, so we have to choose a gray colormap in the

imshow() method call so the image colorspace is properly displayed in the
plot window. axis('off') is used to turn the axes off. The title() method
is used to specify the title of the image. The output is shown in Figure 6-3.

Figure 6-3.  Customizing Matplotlib plot2

2
 Image provided by Hillebrand Steve, U.S. Fish and Wildlife Service (This file is
in public domain, not copyrighted, no rights reserved, free for any use.) License:
https://commons.wikimedia.org/wiki/File:Accent_to_the_top.jpg

121
Chapter 6 Introduction to the Scientific Python Ecosystem

We can use imshow() to push multiple images to an image grid in the

plot window, as shown in Listing 6-6.

Listing 6-6.  prog06.py

import scipy.misc as misc

import matplotlib.pyplot as plt
img1 = misc.face()
img2 = misc.ascent()
titles = ['face', 'ascent']
images = [img1, img2]
plt.subplot(1, 2, 1)
plt.imshow(images[0])
plt.axis('off')
plt.title(titles[0])
plt.subplot(1, 2, 2)
plt.imshow(images[1], cmap='gray')
plt.axis('off')
plt.title(titles[1])
plt.show()

We have used the subplot() method before, with imshow(). The first
two arguments in the subplot() method specify the dimensions of the
grid, and the third argument specifies the position of the image in the grid.
The numbering of the images in the grid starts from the top-left edge. The
top-left position is the first position, the next position is the second one,
and so on. The result is shown in Figure 6-4.

122
 Chapter 6 Introduction to the Scientific Python Ecosystem

Figure 6-4.  Multiple images in a single plot

Image Channels
You can separate image channels of a multi-channel image. The code for
that process is shown in Listing 6-7.

Listing 6-7.  prog07.py

import scipy.misc as misc

import matplotlib.pyplot as plt
img = misc.face()
r = img[:, :, 0]

123
Chapter 6 Introduction to the Scientific Python Ecosystem

g = img[:, :, 1]
b = img[:, :, 2]
titles = ['face', 'Red', 'Green', 'Blue']
images = [img, r, g, b]
plt.subplot(2, 2, 1)
plt.imshow(images[0])
plt.axis('off')
plt.title(titles[0])
plt.subplot(2, 2, 2)
plt.imshow(images[1], cmap='Reds')
plt.axis('off')
plt.title(titles[1])
plt.subplot(2, 2, 3)
plt.imshow(images[2], cmap='Greens')
plt.axis('off')
plt.title(titles[2])
plt.subplot(2, 2, 4)
plt.imshow(images[3], cmap='Blues')
plt.axis('off')
plt.title(titles[3])
plt.show()

The result of the code in Listing 6-7 is shown in Figure 6-5.

124
 Chapter 6 Introduction to the Scientific Python Ecosystem

Figure 6-5.  Color channels

We can use the np.dstack() method, which merges all the channels,
to create the original image, as shown in Listing 6-8.

Listing 6-8.  prog08.py

import scipy.misc as misc

import matplotlib.pyplot as plt
import numpy as np
img = misc.face()
r = img[:, :, 0]
g = img[:, :, 1]

125
Chapter 6 Introduction to the Scientific Python Ecosystem

b = img[:, :, 2]
output = np.dstack((r, g, b))
plt.imshow(output)
plt.axis('off')
plt.title('Combined')
plt.show()

Run the code in Listing 6-8 to see the workings of the np.dstack() for
yourself.

 onversion Between PIL Image Objects

C
and NumPy ndarrays
You can use the methods np.asarray() and Image.fromarray() to
convert between PIL images and NumPy ndarrays, as shown in Listing 6-9.

Listing 6-9.  prog09.py

from PIL import Image

import numpy as np
import matplotlib.pyplot as plt
img = Image.open('/home/pi/DIP/Dataset/4.1.07.tiff')
print(type(img))
num_img = np.asarray(img)
print(type(num_img))
img = Image.fromarray(np.uint8(num_img))
print(type(img))

The console output is as follows:

<class 'PIL.TiffImagePlugin.TiffImageFile'>
<class 'numpy.ndarray'>
<class 'PIL.Image.Image'>

126
 Chapter 6 Introduction to the Scientific Python Ecosystem

Summary
In this chapter, we were introduced to the Scientific Python stack and its
constituent libraries, such as NumPy and matplotlib. We also explored the
scipy.misc module for basic image processing and conversion. In the next
chapter, we will start exploring the scipy.ndimage module for more image
processing operations.

127
CHAPTER 7

Transformations
and Measurements
In the previous chapter, we were introduced to the Scientific Python stack.
We learned the basics of NumPy and matplotlib. We explored the useful
modules ndarray and pyplot from NumPy and matplotlib, respectively.
We also learned about the scipy.misc module and how to perform basic
image processing with it. In this chapter, we will further explore the SciPy
library. We will learn to use the scipy.ndimage library for processing
images. We will also explore methods to use for image transformation and
image measurement.
By the end of this chapter, we will be comfortable with performing
operations on images.

Transformations
We studied a few basic transformations that are possible with scipy.misc
in the previous chapter. Here, we will look at few more.

© Ashwin Pajankar 2022 129

A. Pajankar, Raspberry Pi Image Processing Programming,
https://doi.org/10.1007/978-1-4842-8270-0_7
Chapter 7 Transformations and Measurements

Note  Just like with scipy.misc, scipy.ndimage has an imread()

method that serves the same purpose as scipy.misc.imread().
We will be using face() and ascent() throughout the rest of the
book. These are built in library images in SciPy. However, if we want
to use other images in the Dataset directory, we can use imread()
from the scipy.misc or scipy.ndimage modules in SciPy.

Let’s get started with the simple transformation of shifting. The

shift() method accepts as arguments the image and the values to be
applied to the coordinates for shifting. An example is shown in Listing 7-1.

Listing 7-1.  prog01.py

import scipy.misc as misc

import scipy.ndimage as ndi
import matplotlib.pyplot as plt
img = misc.ascent()
output = ndi.shift(img, 32.0)
plt.imshow(output, cmap='gray')
plt.title('Shift Demo')
plt.axis('off')
plt.show()

Figure 7-1 shows the output.

130
 Chapter 7 Transformations and Measurements

Figure 7-1.  Shift operation demo

We can also zoom in on the image using the zoom() method. We

have to pass the image and the scale of zooming for each of the axes as
arguments to the method. Listing 7-2 shows an example of this operation.

Listing 7-2.  prog02.py

import scipy.misc as misc

import scipy.ndimage as ndi
import matplotlib.pyplot as plt
img = misc.ascent()
plt.imshow(ndi.zoom(img, [5, 3]), cmap='gray')

131
Chapter 7 Transformations and Measurements

plt.title('Zoom Demo')
plt.axis('off')
plt.show()

The output of Listing 7-2 is shown in Figure 7-2.

Figure 7-2.  Zoom demo

We can also use the function rotate() to rotate an image. Listing 7-3
shows several ways to use this function.

132
 Chapter 7 Transformations and Measurements

Listing 7-3.  prog03.py

import scipy.misc as misc

import scipy.ndimage as ndi
import matplotlib.pyplot as plt
img = misc.ascent()
plt.subplot(1, 3, 1)
plt.imshow(img, cmap='gray')
plt.title('Original')
plt.axis('off')
plt.subplot(1, 3, 2)
plt.imshow(ndi.rotate(img, 45, reshape=False), cmap='gray')
plt.title('Not reshaped')
plt.axis('off')
plt.subplot(1, 3, 3)
plt.imshow(ndi.rotate(img, 45, reshape=True), cmap='gray')
plt.title('Reshaped')
plt.axis('off')
plt.show()

The output is shown in Figure 7-3.

133
Chapter 7 Transformations and Measurements

Figure 7-3.  Rotation demo

We can also apply the shift geometric transformation in another way

by defining a custom function. Listing 7-4 demonstrates this.

Listing 7-4.  prog04.py

import scipy.misc as misc

import scipy.ndimage as ndi
import matplotlib.pyplot as plt

134
 Chapter 7 Transformations and Measurements

def shift_func(output_coords,
               x_shift=128,
               y_shift=128):
    return (output_coords[0] - x_shift,
            output_coords[1] - y_shift)

img = misc.ascent()
plt.subplot(1, 2, 1)
plt.imshow(img, cmap='gray')
plt.title('Original')
plt.subplot(1, 2, 2)
plt.imshow(ndi.geometric_transform(img, shift_func),
cmap='gray')
plt.title('Shift')
plt.show()

As seen in Listing 7-4, we pass a customized function for a geometric

transformation. The output is as shown in Figure 7-4.

135
Chapter 7 Transformations and Measurements

Figure 7-4.  Shifting with geometric transformation

Let’s work with an affine transformation next. An affine transformation

is a geometric transformation that preserves the parallelism between lines
in the output. Listing 7-5 demonstrates just such a transformation.

Listing 7-5.  Affine Transformation

import scipy.misc as misc

import scipy.ndimage as ndi
import matplotlib.pyplot as plt
img = misc.ascent()
plt.subplot(1, 3, 1)

136
 Chapter 7 Transformations and Measurements

plt.imshow(img, cmap='gray')
plt.title('Original')
T = [[1, 0, 128], [0, 1, 128], [0, 0, 1]]
plt.subplot(1, 3, 2)
plt.imshow(ndi.affine_transform(img, T), cmap='gray')
plt.title('Translation')
S = [[0.5, 0, 0], [0, 0.5, 0], [0, 0, 1]]
plt.subplot(1, 3, 3)
plt.imshow(ndi.affine_transform(img, S), cmap='gray')
plt.title('Scaling')
plt.show()

In an affine transformation, we have to define a transformation matrix

and then apply it to the image or geometric shape. In Listing 7-5, we apply
scaling and translation (another term for shifting) operations to the image.
Figure 7-5 shows the output.

Figure 7-5.  Affine transformations

We can also apply a spline transformation to an image. We can apply

them to an axis individually or to the whole image, as shown in Listing 7-6.

137
Chapter 7 Transformations and Measurements

Listing 7-6.  Spline Transformation

import scipy.misc as misc

import numpy as np
import scipy.ndimage as ndi
import matplotlib.pyplot as plt
img = misc.ascent()
sp_filter_axis_0 = ndi.spline_filter1d(img, axis=0)
sp_filter_axis_1 = ndi.spline_filter1d(img, axis=1)
sp_filter = ndi.spline_filter(img, order=3)
plt.subplot(2, 2, 1)
plt.imshow(img, cmap='gray')
plt.title('Original')
plt.subplots_adjust(hspace=0.5, wspace=0.5)
plt.subplot(2, 2, 2)
plt.imshow(sp_filter_axis_0, cmap='gray')
plt.title('Spline Filter 1D along X-Axis')
plt.subplot(2, 2, 3)
plt.imshow(sp_filter_axis_1, cmap='gray')
plt.title('Spline Filter 1D along Y-Axis')
plt.subplot(2, 2, 4)
plt.imshow(sp_filter, cmap='gray')
plt.title('Spline Filter')
plt.show()

The output is shown in Figure 7-6.

138
 Chapter 7 Transformations and Measurements

Figure 7-6.  Spline transformations

Measurements
Let’s work with measurements. The first one is the histogram, which is
a visual representation of the frequency distribution of pixel intensities.
Listing 7-7 shows a demonstration of a histogram.

Listing 7-7.  prog07.py

import scipy.misc as misc
import scipy.ndimage as ndi
import matplotlib.pyplot as plt

139
Chapter 7 Transformations and Measurements

img = misc.face()
hist = ndi.histogram(img, 0, 255, 256)
plt.plot(hist, 'k')
plt.title('Face Histogram')
plt.grid(True)
plt.show()

We pass the image, the lowest and highest limits of the bins, and the
number of bins as arguments to the function that computes the histogram.
The output is shown in Figure 7-7.

Figure 7-7.  Histogram

140
 Chapter 7 Transformations and Measurements

Let’s compute a few more measurements, as shown in Listing 7-8.

Listing 7-8.  prog08.py

import scipy.misc as misc

import scipy.ndimage as ndi
img = misc.ascent()
print(ndi.minimum(img))
print(ndi.minimum_position(img))
print(ndi.maximum(img))
print(ndi.maximum_position(img))
print(ndi.extrema(img))

The output is as follows:

0
(201, 268)
255
(190, 265)
(0, 255, (201, 268), (190, 265))

We can compute a few statistical properties as well, as shown in

Listing 7-9.

Listing 7-9.  prog09.py

import scipy.misc as misc

import scipy.ndimage as ndi
img = misc.ascent()
print(ndi.sum(img))
print(ndi.mean(img))
print(ndi.median(img))
print(ndi.variance(img))
print(ndi.standard_deviation(img))

141
Chapter 7 Transformations and Measurements

The output is as follows:

22932324
87.47987365722656
80.0
2378.9479362999555
48.774459877070456

In addition, we can compute the center of mass, as demonstrated in

Listing 7-10.

Listing 7-10.  prog10.py

import scipy.misc as misc

import scipy.ndimage as ndi
img = misc.ascent()
print(ndi.center_of_mass(img))

The output is as follows:

(259.9973423539629, 254.6090727219797)

Computing the labels in the image can also be achieved, as

demonstrated in Listing 7-11.

Listing 7-11.  prog11.py

import scipy.misc as misc

import scipy.ndimage as ndi
import numpy as np
img = misc.face()
a = np.zeros((6,6), dtype=int)
a[2:4, 2:4] = 1
a[4, 4] = 1
a[:2, :3] = 2
a[0, 5] = 3

142
 Chapter 7 Transformations and Measurements

print(ndi.find_objects(a))
print(ndi.find_objects(a, max_label=2))
print(ndi.find_objects(a, max_label=4))
print(ndi.find_objects(img, max_label=4))
print(ndi.sum_labels(img))

The output is as follows:

[(slice(2, 5, None), slice(2, 5, None)), (slice(0, 2, None),

slice(0, 3, None)), (slice(0, 1, None), slice(5, 6, None))]
[(slice(2, 5, None), slice(2, 5, None)), (slice(0, 2, None),
slice(0, 3, None))]
[(slice(2, 5, None), slice(2, 5, None)), (slice(0, 2, None),
slice(0, 3, None)), (slice(0, 1, None), slice(5, 6, None)), None]
[(slice(1, 768, None), slice(7, 1024, None), slice(0, 3, None)),
(slice(1, 768, None), slice(8, 1024, None), slice(0, 3, None)),
(slice(1, 768, None), slice(0, 1024, None), slice(0, 3, None)),
(slice(0, 768, None), slice(1, 1023, None), slice(0, 3, None))]
259906521

Summary
In this chapter, we explored methods for transformations and
measurements. We studied the shift and zoom transformations and
calculated the histogram of a grayscale image. We also calculated statistical
information about the images.
In the next chapter, we will study image kernels and filters, their types,
and their applications in image enhancement in detail.

143
CHAPTER 8

Filters
In the previous chapter, we learned the functionality found in the
scipy.ndimage module of the SciPy library. We learned how to apply
transformations such as shift and zoom on an image. We also learned how
to obtain statistical information about an image. We saw how to compute
and plot the histogram for an image.
In Chapter 5, we studied image filters using the Pillow library. In this
chapter, we will study in detail the theory behind those filters. We will also
see the types of filters and kernels used in image processing, as well as the
applications of those filters.
The following are topics covered in this chapter:

• Kernels

• Low-pass filters
• High-pass filters

• Fourier filters

By the end of this chapter, we will be comfortable with SciPy filters and
their applications.

© Ashwin Pajankar 2022 145

A. Pajankar, Raspberry Pi Image Processing Programming,
https://doi.org/10.1007/978-1-4842-8270-0_8
Chapter 8 Filters

Kernels, Convolution, and Correlation

A kernel (also known as a convolution matrix or a mask) is a two-­
dimensional matrix used to perform image processing. We can use kernels
for operations like convolution and correlation.
Convolution adds each element of the matrix to its local neighbors,
weighted by the kernel. Convolution, depending on the kernels used, can
produce a variety of results, such as blurring, sharpening, embossing, and
edge detection.
While using convolution on matrices, we can use various techniques
to handle edges, such as extend, wrap, mirror, crop, kernel crop, constant,
and avoid overlap.
Let’s demonstrate the functions convolve1d() and convolve() on
one-dimensional and two-dimensional matrices, as shown in Listing 8-1.

Listing 8-1.  prog00.py

import scipy.ndimage as ndi

import numpy as np
print(ndi.convolve1d([0, 1, 2, 3, 4, 5, 6],
                     weights=[1, 2, 3]))
a = np.array([[0, 1, 2, 3],
              [4, 5, 6, 7],
              [8, 9, 10, 11],
              [12, 13, 14, 15]])
k = np.array([[0 ,1, 0],[1, 1, 1],[0, 1, 0]])
print(ndi.convolve(a, k, mode='nearest'))

The output is as follows:

[ 1  4 10 16 22 28 33]
[[ 5  9 14 18]
[21 25 30 34]

146
 Chapter 8 Filters

[41 45 50 54]
[57 61 66 70]]

We can also apply the convolution operation to image matrices. The

tutorial at https://setosa.io/ev/image-­kernels discusses various
kernels and their effects on images. We can use the function convolve2d()
from SciPy’s signal module. Listing 8-2 provides a demonstration.

Listing 8-2.  prog01.py

import scipy.ndimage as ndi

import scipy.signal
import numpy as np
import scipy.misc as misc
import matplotlib.pyplot as plt
img = misc.ascent()
k = np.ones((11, 11), np.uint8)/121
c1 = ndi.convolve(img, k, mode='nearest')
c2 = scipy.signal.convolve2d(img, k)
print(c1)
print(c2)
plt.subplot(1, 3, 1)
plt.imshow(img, cmap='gray')
plt.title('Original')
plt.subplot(1, 3, 2)
plt.imshow(c1, cmap='gray')
plt.title('Ndimage Convolution')
plt.subplot(1, 3, 3)
plt.imshow(c2, cmap='gray')
plt.title('Signal Convolution')
plt.show()

It produces the output shown in Figure 8-1.

147
Chapter 8 Filters

Figure 8-1.  Convolution demo

Convolution is a mathematical way of generating a single output by

combining two inputs. Correlation compares two signals. Listing 8-3 is a
demonstration of the correlation operation,

Listing 8-3.  prog02.py

import scipy.ndimage as ndi

import numpy as np
print(ndi.correlate1d([0, 1, 2, 3, 4, 5, 6],
                     weights=[1, 2, 3]))
a = np.array([[0, 1, 2, 3],
              [4, 5, 6, 7],
              [8, 9, 10, 11],
              [12, 13, 14, 15]])
k = np.array([[0 ,1, 0],[1, 1, 1],[0, 1, 0]])
print(ndi.correlate(a, k, mode='nearest'))

The output is as follows:

[ 3 8 14 20 26 32 35]
[[ 5 9 14 18]
[21 25 30 34]
[41 45 50 54]
[57 61 66 70]]

148
 Chapter 8 Filters

Let’s apply correlation to an image. We will (just like Listing 8-2)

use two different methods to compute correlation, as demonstrated in
Listing 8-4.

Listing 8-4.  prog03.py

import scipy.ndimage as ndi

import scipy.signal
import numpy as np
import scipy.misc as misc
import matplotlib.pyplot as plt
img = misc.ascent()
k = np.ones((11, 11), np.uint8)/121
c1 = ndi.correlate(img, k, mode='nearest')
c2 = scipy.signal.correlate2d(img, k)
print(c1)
print(c2)
plt.subplot(1, 3, 1)
plt.imshow(img, cmap='gray')
plt.title('Original')
plt.subplot(1, 3, 2)
plt.imshow(c1, cmap='gray')
plt.title('Ndimage Correlation')
plt.subplot(1, 3, 3)
plt.imshow(c2, cmap='gray')
plt.title('Signal Correlation')
plt.show()

The output is as shown in Figure 8-2.

149
Chapter 8 Filters

Figure 8-2.  Correlation demo

Low-Pass Filters
We can use various kernels with the convolution operation to create filters.
Filters are used to enhance an image or to perform operations on an
image. The filters that remove the high-frequency data from a signal (or
an image, for that matter) are known as low-pass filters. There are many
applications of low-pass filters. Let’s look at them now.

Blurring
We can blur images with blurring kernels. SciPy comes with many routines
that use blurring kernels with convolution. Listing 8-5 demonstrates a
Gaussian filter’s being applied to an image.

Listing 8-5.  prog04.py

import scipy.misc as misc

import scipy.ndimage as ndi
import matplotlib.pyplot as plt
img = misc.face()
output1 = ndi.gaussian_filter(img, sigma=3)
output2 = ndi.gaussian_filter(img, sigma=5)

150
 Chapter 8 Filters

output3 = ndi.gaussian_filter(img, sigma=7)

output = [output1, output2, output3]
titles = ['Sigma = 3', 'Sigma = 5',
          'Sigma = 7']
for i in range(3):
      plt.subplot(1, 3, i+1)
      plt.imshow(output[i], cmap='gray')
      plt.title(titles[i])
      plt.axis('off')
plt.show()

The routine for the Gaussian filter accepts the image and the value for
sigma. The output is as shown in Figure 8-3.

Figure 8-3.  Gaussian filter demo

A uniform filter replaces the value of a pixel by the mean value of an area
centered at the pixel. Let’s see a uniform filter, as demonstrated in Listing 8-6.

Listing 8-6.  prog05.py

import scipy.misc as misc

import scipy.ndimage as ndi
import matplotlib.pyplot as plt
img = misc.face()
output1 = ndi.uniform_filter(img, size=19)
output2 = ndi.uniform_filter(img, size=25)
output3 = ndi.uniform_filter(img, size=31)

151
Chapter 8 Filters

output = [output1, output2, output3]

titles = ['Size = 19', 'Size = 25', 'Size = 31']
for i in range(3):
        plt.subplot(1, 3, i+1)
        plt.imshow(output[i], cmap='gray')
        plt.title(titles[i])
        plt.axis('off')
plt.show()

The routine for the uniform filter accepts the image and the size of the
filter. The output is as shown in Figure 8-4.

Figure 8-4.  Uniform filter demo

We can also have a one-dimensional Gaussian filter, as demonstrated

in Listing 8-7.

Listing 8-7.  prog06.py

import scipy.misc as misc

import scipy.ndimage as ndi
import matplotlib.pyplot as plt
img = misc.ascent()
y3 = ndi.gaussian_filter1d(img, sigma=3)
y6 = ndi.gaussian_filter1d(img, sigma=6)
plt.subplot(1, 3, 1)
plt.imshow(img, cmap='gray')
plt.title('Original Image')

152
 Chapter 8 Filters

plt.subplot(1, 3, 2)
plt.imshow(y3, cmap='gray')
plt.title('Sigma = 3')
plt.subplot(1, 3, 3)
plt.imshow(y6, cmap='gray')
plt.title('Sigma = 3')
plt.show()

The routine for the one-dimensional Gaussian filter accepts the image
and the value of Sigma. The output is as shown in Figure 8-5.

Figure 8-5.  One-dimensional Gaussian filter demo

We can even apply a one-dimensional uniform filter to an image, as

demonstrated in Listing 8-8.

Listing 8-8.  prog07.py

import scipy.misc as misc

import scipy.ndimage as ndi
import matplotlib.pyplot as plt
img = misc.ascent()
y9 = ndi.uniform_filter1d(img, size=9)
y12 = ndi.uniform_filter1d(img, size=12)
plt.subplot(1, 3, 1)
plt.imshow(img, cmap='gray')

153
Chapter 8 Filters

plt.title('Original Image')
plt.subplot(1, 3, 2)
plt.imshow(y9, cmap='gray')
plt.title('Size = 9')
plt.subplot(1, 3, 3)
plt.imshow(y12, cmap='gray')
plt.title('Size = 12')
plt.show()

The routine for the one-dimensional uniform filter accepts the image
and the size of the filter. The output is as shown in Figure 8-6.

Figure 8-6.  One-dimensional uniform filter demo

A percentile filter computes the n-percentile value of neighborhood

pixel and applies it to the center pixel. We can apply a percentile filter to an
image, as demonstrated in Listing 8-9.

Listing 8-9.  prog08.py

import scipy.misc as misc

import scipy.ndimage as ndi
import matplotlib.pyplot as plt
img = misc.ascent()
out1 = ndi.percentile_filter(img, size=10,
                             percentile=10)

154
 Chapter 8 Filters

out2 = ndi.percentile_filter(img, size=45,

                             percentile=65)
plt.subplot(1, 3, 1)
plt.imshow(img, cmap='gray')
plt.title('Original Image')
plt.subplot(1, 3, 2)
plt.imshow(out1, cmap='gray')
plt.title('percentile=10')
plt.subplot(1, 3, 3)
plt.imshow(out2, cmap='gray')
plt.title('percentile=65')
plt.show()

The output is as shown in Figure 8-7.

Figure 8-7.  Percentile filter demo

Rank filters are non-linear filters. A rank filter uses local gray-level
ordering to compute the output. We can even apply a rank filter to an
image, as demonstrated in Listing 8-10.

Listing 8-10.  prog09.py

import scipy.misc as misc
import scipy.ndimage as ndi
import matplotlib.pyplot as plt

155
Chapter 8 Filters

img = misc.ascent()
out1 = ndi.rank_filter(img, size=10, rank=10)
out2 = ndi.rank_filter(img, size=45, rank=65)
plt.subplot(1, 3, 1)
plt.imshow(img, cmap='gray')
plt.title('Original Image')
plt.subplot(1, 3, 2)
plt.imshow(out1, cmap='gray')
plt.title('rank=10')
plt.subplot(1, 3, 3)
plt.imshow(out2, cmap='gray')
plt.title('rank=65')
plt.show()

The output is as shown in Figure 8-8.

Figure 8-8.  Rank filter demo

Noise Reduction
We can use low-pass filters for noise reduction. In the next two examples
(Listing 8-11 and Listing 8-12), we will create artificial noise with NumPy
routines and add it to the test image. Listing 8-11 shows how to reduce
noise with the Gaussian filter.

156
 Chapter 8 Filters

Listing 8-11.  prog10.py

import scipy.misc as misc

import scipy.ndimage as ndi
import numpy as np
import matplotlib.pyplot as plt
img = misc.ascent()
noisy = img + 0.8 * img.std() * np.random.random(img.shape)
output1 = ndi.gaussian_filter(noisy, sigma=1)
output2 = ndi.gaussian_filter(noisy, sigma=3)
output3 = ndi.gaussian_filter(noisy, sigma=5)
output = [noisy, output1, output2, output3]
titles = ['Noisy', 'Sigma = 1', 'Sigma = 3', 'Sigma = 5']
for i in range(4):
      plt.subplot(2, 2, i+1)
      plt.imshow(output[i], cmap='gray')
      print(output[i])
      plt.title(titles[i])
      plt.axis('off')
plt.show()

The output is as shown in Figure 8-9.

157
Chapter 8 Filters

Figure 8-9.  Gaussian filter for noise reduction

Similarly, we can apply a median filter on a noisy image, as

demonstrated in Listing 8-12.

Listing 8-12.  prog11.py

import scipy.misc as misc

import scipy.ndimage as ndi
import numpy as np
import matplotlib.pyplot as plt
img = misc.ascent()
noisy = img + 0.8 * img.std() * np.random.random(img.shape)
output1 = ndi.median_filter(noisy, 3)
output2 = ndi.median_filter(noisy, 7)
output3 = ndi.median_filter(noisy, 9)
output = [noisy, output1, output2, output3]

158
 Chapter 8 Filters

titles = ['Noisy', 'Size = 3', 'Size = 7', 'Size = 9']

for i in range(4):
      plt.subplot(2, 2, i+1)
      plt.imshow(output[i], cmap='gray')
      plt.title(titles[i])
      plt.axis('off')
plt.show()

The output is as shown in Figure 8-10.

Figure 8-10.  Median filter for noise reduction

Minimum and maximum filters can also be applied to a noisy image,

as demonstrated in Listing 8-13.

Listing 8-13.  prog12.py

import scipy.misc as misc

import scipy.ndimage as ndi

159
Chapter 8 Filters

import numpy as np
import matplotlib.pyplot as plt
img = misc.ascent()
output1 = ndi.minimum_filter1d(img, 9)
output2 = ndi.minimum_filter(img, 9)
output3 = ndi.maximum_filter1d(img, 9)
output4 = ndi.maximum_filter(img, 9)
output = [output1, output2, output3, output4]
titles = ['Minimum 1D', 'Minimum',
          'Maximum 1D', 'Maximum']
for i in range(4):
      plt.subplot(2, 2, i+1)
      plt.imshow(output[i], cmap='gray')
      plt.title(titles[i])
      plt.axis('off')
plt.show()

The output is as shown in Figure 8-11.

160
 Chapter 8 Filters

Figure 8-11.  Minimum and maximum filters

High-Pass Filters
High-pass filters work in the exact opposite way to the low-pass filters.
They allow high-frequency signals to pass through them. They also use
kernels and convolutions. However, the kernel used for high-pass filters is
different than that used in low-pass filters. When we apply high-pass filters
to an image, it sharpens the image (which is opposite of blurring) and
highlights edges.
Listing 8-14 demonstrates the Prewitt filter to detect edges in the
test image.

161
Chapter 8 Filters

Listing 8-14.  prog13.py

import scipy.misc as misc

import scipy.ndimage as ndi
import matplotlib.pyplot as plt
img = misc.ascent()
filtered = ndi.prewitt(img)
output = [img, filtered]
titles = ['Original', 'Filtered']
for i in range(2):
      plt.subplot(1, 2, i+1)
      plt.imshow(output[i], cmap='gray')
      plt.title(titles[i])
      plt.axis('off')
plt.show()

The output is as shown in Figure 8-12.

Figure 8-12.  Prewitt filter

We can see the highlighted edges in Figure 8-12.

162
 Chapter 8 Filters

We can also apply Sobel filters to axes separately and to the entire
image. It will detect horizontal edges, vertical edges, and edges on both the
axes. Listing 8-15 generates a test image and applies Sobel filters.

Listing 8-15.  prog14.py

import numpy as np
import scipy.ndimage as ndi
import matplotlib.pyplot as plt
img = np.zeros((516, 516))
img[128:-128, 128:-128] = 1
img = ndi.gaussian_filter(img, 8)
rotated = ndi.rotate(img, -20)
noisy = rotated + 0.09 * np.random.random(rotated.shape)
sx = ndi.sobel(noisy, axis=0)
sy = ndi.sobel(noisy, axis=1)
sob = np.hypot(sx, sy)
titles = ['Original', 'Rotated', 'Noisy',
          'Sobel (X-axis)', 'Sobel (Y-axis)', 'Sobel']
output = [img, rotated, noisy, sx, sy, sob]
for i in range(6):
      plt.subplot(2, 3, i+1)
      plt.imshow(output[i])
      plt.title(titles[i])
      plt.axis('off')
plt.show()

The output is as shown in Figure 8-13.

163
Chapter 8 Filters

Figure 8-13.  Sobel filters

Finally, Listing 8-16 demonstrates the Laplacian, Gaussian gradient

magnitude, and Gaussian Laplace filters.

Listing 8-16.  prog15.py

import scipy.misc as misc

import scipy.ndimage as ndi
import matplotlib.pyplot as plt
img = misc.ascent()
output1 = ndi.laplace(img)
output2 = ndi.gaussian_gradient_magnitude(img,
                                          sigma=3)
output3 = ndi.gaussian_laplace(img, sigma=3)
titles = ['Original', 'Laplacian',
          'GGM', 'Gaussian Laplace']
output = [img, output1,
          output2, output3]
for i in range(4):

164
 Chapter 8 Filters

      plt.subplot(2, 2, i+1)
      plt.imshow(output[i],
                   cmap='gray')
      plt.title(titles[i])
      plt.axis('off')
plt.show()

The output is as shown in Figure 8-14.

Figure 8-14.  Laplacian, Gaussian gradient magnitude, and

Gaussian Laplace filters

Fourier Filters
Fourier filters work in the frequency domain. They compute the fourier
transform of the image, manipulate frequencies, and then finally compute the
inverse fourier transform to produce the final output. We can apply various
Fourier filtersFiltersfourier filters to images, as demonstrated in Listing 8-17.

165
Chapter 8 Filters

Listing 8-17.  prog16.py

import scipy.ndimage as ndi

import scipy.misc as misc
import matplotlib.pyplot as plt
import numpy as np
img = misc.ascent()
noisy = img + 0.09 * img.std() * np.random.random(img.shape)
fe = ndi.fourier_ellipsoid(img, 1)
fg = ndi.fourier_gaussian(img, 1)
fs = ndi.fourier_shift(img, 1)
fu = ndi.fourier_uniform(img, 1)
titles = ['Original', 'Noisy',
          'Fourier Ellipsoid', 'Fourier Gaussian',
          'Fourier Shift', 'Fourier Uniform']
output = [img, noisy, fe, fg, fs, fu]
for i in range(6):
      plt.subplot(2, 3, i+1)
      plt.imshow(np.float64(output[i]), cmap='gray')
      plt.title(titles[i])
      plt.axis('off')
plt.show()

The output is as shown in Figure 8-15.

166
 Chapter 8 Filters

Figure 8-15.  Various Fourier filters

Summary
In this chapter, we were introduced to myriad filters. We saw their types
and looked at their applications. The image-filtering topic is too vast to be
completely explored in a single chapter.
In the next chapter, we will cover morphological operators, image
thresholding, and basic segmentation.

167
CHAPTER 9

Morphology,
Thresholding,
and Segmentation
In the previous chapter, we studied the theory behind image filters, along
with the types and practical applications of filters used to enhance images.
In this chapter, we are going to study and demonstrate important
concepts in image processing, such as morphology, morphological
operations on images, thresholding, and segmentation. We will study and
demonstrate the following:

• Distance transforms
• Morphology and morphological operations

• Thresholding and segmentation

Distance Transforms
A distance transform is an operation performed on binary images.
Binary images have background elements (zero value – black color)
and foreground elements (white color). A distance transform replaces

© Ashwin Pajankar 2022 169

A. Pajankar, Raspberry Pi Image Processing Programming,
https://doi.org/10.1007/978-1-4842-8270-0_9
Chapter 9 Morphology, Thresholding, and Segmentation

each foreground element with the value of the shortest distance to the
background. scipy.ndimage has three methods for computing the
distance transform of a binary image. The code in Listing 9-1 illustrates
how a distance transform can be used practically to generate test images.

Listing 9-1.  prog01.py

import scipy.ndimage as ndi

import matplotlib.pyplot as pltimport numpy as np
img = np.zeros((32, 32))
img[8:-8, 8:-8] = 1
print(img)
dist1 = ndi.distance_transform_bf(img)
dist2 = ndi.distance_transform_cdt(img)
dist3 = ndi.distance_transform_edt(img)
output = [img, dist1, dist2, dist3]
titles = ['Original', 'Brute Force', 'Chamfer', 'Euclidean']
for i in range(4):
      print(output[i])
      plt.subplot(2, 2, i+1)
      plt.imshow(output[i],
               interpolation='nearest')
      plt.title(titles[i])
      plt.axis('off')
plt.show()

The code shown in Listing 9-1 computes the distance transform by

using a brute-force algorithm, a Chamfer-type algorithm, and Euclidean
methods, respectively. We are going to use distance transforms to generate
the test images in this chapter. The output is shown in Figure 9-1.

170
 Chapter 9 Morphology, Thresholding, and Segmentation

Figure 9-1.  Distance transform

Morphology and Morphological Operations

Morphology is the study of shapes and forms. The morphological study
of images deals with shapes rather than with the values of the pixels.
Morphological operations are usually performed on binary images. Let’s
take a look at a few concepts related to morphological operations.

171
Chapter 9 Morphology, Thresholding, and Segmentation

Structuring Element
A structuring element is a matrix that’s used to interact with a given binary
image. It comes in various shapes, like a ball, a ring, or a line. It can come
in many shapes, like a 3 x 3 or a 7 x 7 matrix. Larger structuring elements
take more time for computation. A simple structuring element can be
defined as a unity matrix of odd sizes. np.ones((3, 3)) is an example
of this.

Binary Morphological Operations

Let’s briefly look at various morphological operations. Dilation causes the
expansion of the shapes in an input image. Erosion causes shrinkage in the
shape in an input image. Opening is the dilation of erosion. Closing is the
erosion of dilation.
It is difficult to understand these concepts just by reading about
them. Take a look at the example in Listing 9-2, which demonstrates these
concepts on a binary image.

Listing 9-2.  prog02.py

import matplotlib.pyplot as plt

import scipy.ndimage as ndi
import numpy as np
img = np.zeros((16, 16))
img[4:-4, 4:-4] = 1
print(img)
erosion = ndi.binary_erosion(img).astype(img.dtype)
dilation = ndi.binary_dilation(img).astype(img.dtype)
opening = ndi.binary_opening(img).astype(img.dtype)
closing = ndi.binary_closing(img).astype(img.dtype)
output = [img, erosion, dilation,

172
 Chapter 9 Morphology, Thresholding, and Segmentation

          opening, closing]
titles = ['Original', 'Erosion',
          'Dilation', 'Opening',
          'Closing']
for i in range(5):
      print(output[i])
      plt.subplot(1, 5, i+1)
      plt.imshow(output[i],
               interpolation='nearest')
      plt.title(titles[i])
      plt.axis('off')
plt.show()

The code example in Listing 9-2 generates a binary image and applies
all the binary morphological operations to it. The output is shown in
Figure 9-2.

Figure 9-2.  Morphological operations on a binary image

Another important operation is binary_fill_holes(). It is used to fill

the gaps in the binary image, as shown in Listing 9-3.

Listing 9-3.  prog03.py

import matplotlib.pyplot as plt

import scipy.ndimage as ndi
import numpy as np
img = np.ones((32, 32))
x, y = (32*np.random.random((2, 20))).astype(np.int)

173
Chapter 9 Morphology, Thresholding, and Segmentation

img[x, y] = 0
noise_removed = ndi.binary_fill_holes(img).astype(int)
output = [img, noise_removed]
titles = ['Original', 'Noise Removed']
for i in range(2):
      print(output[i])
      plt.subplot(1, 2, i+1)
      plt.imshow(output[i],
               interpolation='nearest')
      plt.title(titles[i])
      plt.axis('off')
plt.show()

The code in Listing 9-3 generates a 32 x 32 square matrix image with

all the values as 1 (white value). Then it randomly sets a few pixels to 0 (the
dark value). The dark pixels can be considered holes in the image, which
can then be removed using binary_fill_holes().
The output is shown in Figure 9-3.

Figure 9-3.  Noise removal with a morphological operation

174
 Chapter 9 Morphology, Thresholding, and Segmentation

Earlier, we discussed binary structure. Let’s apply the morphological

operations on an image using a binary structure. Listing 9-4
demonstrates this.

Listing 9-4.  prog04.py

import matplotlib.pyplot as plt

import scipy.ndimage as ndi
import numpy as np
img = np.zeros((16, 16))
img[4:-4, 4:-4] = 1
struct = ndi.generate_binary_structure(2, 1)
print(struct)
erosion = ndi.binary_erosion(img, struct).astype(img.dtype)
dilation = ndi.binary_dilation(img, struct).astype(img.dtype)
opening = ndi.binary_opening(img, struct).astype(img.dtype)
closing = ndi.binary_closing(img, struct).astype(img.dtype)
output = [img, erosion, dilation,
          opening, closing]
titles = ['Original', 'Erosion',
          'Dilation', 'Opening',
          'Closing']
for i in range(5):
      print(output[i])
      plt.subplot(1, 5, i+1)
      plt.imshow(output[i],
               interpolation='nearest')
      plt.title(titles[i])
      plt.axis('off')
plt.show()

The output is as shown in Figure 9-4.

175
Chapter 9 Morphology, Thresholding, and Segmentation

Figure 9-4.  Morphological operations with structuring element

We can apply a hit or miss operation with and without a structuring

element, as demonstrated in Listing 9-5.

Listing 9-5.  prog05.py

import matplotlib.pyplot as plt

import scipy.ndimage as ndi
import numpy as np
img = np.zeros((7, 7),
               dtype=int)
img[1, 1] = img[2:4, 2:4] = img[4:6, 4:6] = 1
struct = np.array([[1, 0, 0],
                   [0, 1, 1],
                   [0, 1, 1]])
print(struct)
hit_or_miss1 = ndi.binary_hit_or_miss(img).astype(int)
hit_or_miss2 = ndi.binary_hit_or_miss(img, structure1=struct).
astype(int)
output = [img, hit_or_miss1, hit_or_miss2]
titles = ['Original',
          'Hit or Miss 1',
          'Hit or Miss 2']
for i in range(3):
      print(output[i])
      plt.subplot(1, 3, i+1)

176
 Chapter 9 Morphology, Thresholding, and Segmentation

      plt.imshow(output[i],
                   interpolation='nearest')
      plt.title(titles[i])
      plt.axis('off')
plt.show()

The output is as shown in Figure 9-5.

Figure 9-5.  Binary hit or miss operation

We can even demonstrate binary propagation, as shown in Listing 9-6.

Listing 9-6.  prog06.py

import matplotlib.pyplot as plt

import scipy.ndimage as ndi
import numpy as np
n = 10
img = np.zeros((n, n),
               dtype=np.uint8)
img[2, 2] = 1
mask = np.zeros((n, n),
                dtype=np.uint8)
mask[1:4, 1:4] = mask[4, 4] = mask[6:8, 6:8] = 1
struct = np.ones((3, 3))
o1=ndi.binary_propagation(img, mask=mask).astype(int)

177
Chapter 9 Morphology, Thresholding, and Segmentation

o2=ndi.binary_propagation(img, mask=mask,
                          structure=struct).astype(int)
output = [img, o1, o2]
titles = ['Original',
          'Output 1',
          'Output 2']
for i in range(3):
      print(output[i])
      plt.subplot(1, 3, i+1)
      plt.imshow(output[i],
               interpolation='nearest')
      plt.title(titles[i])
      plt.axis('off')
plt.show()

The output is as shown in Figure 9-6.

Figure 9-6.  Binary propagation

Grayscale Morphological Operations

There is a set of grayscale morphological operations. The code shown
in Listing 9-7 generates eight random values and assigns them to pixels
on a completely dark (zero value) background. Then it uses the grey_
dilation() method to dilate them.

178
 Chapter 9 Morphology, Thresholding, and Segmentation

Listing 9-7.  prog07.py

import matplotlib.pyplot as plt

import scipy.ndimage as ndi
import numpy as np
img = np.zeros((64, 64))
x, y = (63*np.random.random((2, 8))).astype(np.int)
img[x, y] = np.arange(8)
dilation = ndi.grey_dilation(img, size=(5, 5),
                             structure=np.ones((5, 5)))
output = [img, dilation]
titles = ['Original', 'Dilation']
for i in range(2):
      print(output[i])
      plt.subplot(1, 2, i+1)
      plt.imshow(output[i],
                   interpolation='nearest')
      plt.title(titles[i])
      plt.axis('off')
plt.show()

The output is shown in Figure 9-7.

179
Chapter 9 Morphology, Thresholding, and Segmentation

Figure 9-7.  Gray dilation

We are using a structuring element that’s 5 x 5 here.

Note This example uses the random() method. When you will

execute the code in the bundle, the output won’t be exactly the same.

The code shown in Listing 9-8 applies gray dilation and gray erosion
operations to a distance transform.

Listing 9-8.  prog08.py

import matplotlib.pyplot as plt

import scipy.ndimage as ndi
import numpy as np

img = np.zeros((16, 16))

img[4:-4, 4:-4] = 1

img = ndi.distance_transform_bf(img)

180
 Chapter 9 Morphology, Thresholding, and Segmentation

dilation = ndi.grey_dilation(img, size=(3, 3),

                             structure=np.ones((3, 3)))

erosion = ndi.grey_erosion(img, size=(3, 3),

                           structure=np.ones((3, 3)))

output = [img, dilation, erosion]

titles = ['Original', 'Dilation', 'Erosion']

for i in range(3):
      print(output[i])
      plt.subplot(1, 3, i+1)
      plt.imshow(output[i], interpolation='nearest',
cmap='rainbow')
      plt.title(titles[i])
      plt.axis('off')
plt.show()

The code in Listing 9-8 uses a structuring element that’s 3 x 3 for both
operations. The output is shown in Figure 9-8.

Figure 9-8.  Dilation and erosion on a distance transform

181
Chapter 9 Morphology, Thresholding, and Segmentation

Thresholding and Segmentation
This is the final part of the book and it deals with one of the most
important applications of image processing: segmentation. In thresholding
operations, you convert grayscale images to binary (black and white)
images based on the threshold value. The pixels with intensity values
greater than the threshold are assigned white, and the pixels with intensity
values lower than the threshold are assigned a dark value. This is known
as binary thresholding, and is the most basic form of thresholding and
segmentation. An example is shown in Listing 9-9.

Listing 9-9.  prog09.py

import matplotlib.pyplot as plt

import scipy.misc as misc
img = misc.ascent()
thresh = img > 127
output = [img, thresh]
titles = ['Original', 'Thresholding']
for i in range(2):
      plt.subplot(1, 2, i+1)
      plt.imshow(output[i], cmap='gray')
      plt.title(titles[i])
      plt.axis('off')
plt.show()

The code in Listing 9-9 sets the threshold at 127. At the grayscale, a
pixel value of 127 corresponds to the gray color. The resulting thresholded
image is shown in Figure 9-9.

182
 Chapter 9 Morphology, Thresholding, and Segmentation

Figure 9-9.  Binary thresholding

Thresholding is the most basic type of image segmentation. Image

segmentation refers to dividing an image into many regions based on
some property, like colors of pixels, connectivity of the region, and so forth.
You can get the better segments of an image by applying morphological
operations to a thresholded image (see Listing 9-10).

Listing 9-10.  prog10.py

import matplotlib.pyplot as plt

import scipy.misc as misc
import scipy.ndimage as ndi
import numpy as np
img = misc.ascent()
thresh = img > 127
dilated = ndi.binary_dilation(thresh, structure=np.ones((9,
9))).astype(int)
eroded = ndi.binary_erosion(dilated, structure=np.ones((9,
9))).astype(int)
output = [img, thresh, dilated, eroded]

183
Chapter 9 Morphology, Thresholding, and Segmentation

titles = ['Original', 'Thresholding', 'Dilated', 'Eroded and

Segmented']
for i in range(4):
      plt.subplot(2, 2, i+1)
      plt.imshow(output[i], cmap='gray')
      plt.title(titles[i])
      plt.axis('off')
plt.show()

The output is shown in Figure 9-10.

Figure 9-10.  Thresholded and segmented image

The top-hat transformation extracts small elements from image. we

can compute white tophat and black tophat as demonstrated in
Listing 9-11.

184
 Chapter 9 Morphology, Thresholding, and Segmentation

Listing 9-11.  prog11.py

import matplotlib.pyplot as plt

import scipy.misc as misc
import scipy.ndimage as ndi
img = misc.ascent()
struct = ndi.generate_binary_structure(rank=2,
                                       connectivity=3)
o1 = ndi.black_tophat(input=img,
                      structure=struct)
o2 = ndi.white_tophat(input=img,
                      structure=struct)
output = [img, o1, o2]
titles = ['Original', 'Black Tophat',
          'White Tophat']
for i in range(3):
        plt.subplot(1, 3, i+1)
        plt.imshow(output[i],
                   cmap='gray')
        plt.title(titles[i])
        plt.axis('off')
plt.show()

The output is shown in Figure 9-11.

Figure 9-11.  Black and white tophat

185
Chapter 9 Morphology, Thresholding, and Segmentation

We can also iterate over a structure, as demonstrated in Listing 9-12.

Listing 9-12.  prog12.py

import scipy.ndimage as ndi

struct = ndi.generate_binary_structure(2, 1).astype(int)
print(struct)
print(ndi.iterate_structure(struct, 2).astype(int))
print(ndi.iterate_structure(struct, 3).astype(int))

The output is as follows:

[[0 1 0]
[1 1 1]
[0 1 0]]

[[0 0 1 0 0]
[0 1 1 1 0]
[1 1 1 1 1]
[0 1 1 1 0]
[0 0 1 0 0]]

[[0 0 0 1 0 0 0]
[0 0 1 1 1 0 0]
[0 1 1 1 1 1 0]
[1 1 1 1 1 1 1]
[0 1 1 1 1 1 0]
[0 0 1 1 1 0 0]
[0 0 0 1 0 0 0]]

We can also compute the morphological gradient, as demonstrated in

Listing 9-13.

186
 Chapter 9 Morphology, Thresholding, and Segmentation

Listing 9-13.  prog13.py

import matplotlib.pyplot as plt

import scipy.misc as misc
import scipy.ndimage as ndi
img = misc.ascent()
o1 = ndi.morphological_gradient(img, size=(3, 3))
struct = ndi.generate_binary_structure(2, 1)
o2 = ndi.morphological_laplace(img,
                               structure=struct)
output = [img, o1, o2]
titles = ['Original',
          'M Gradient',
          'M Laplace']
for i in range(3):
        plt.subplot(1, 3, i+1)
        plt.imshow(output[i],
                   cmap='gray')
        plt.title(titles[i])
        plt.axis('off')
plt.show()

The output is shown in Figure 9-12.

Figure 9-12.  Morphological gradients

187
Chapter 9 Morphology, Thresholding, and Segmentation

Summary
In this chapter, we studied distance transforms for generating test images.
We then learned about morphology and how to use morphological
operations on images. Morphological operations come in two varieties—
binary and grayscale. We also studied thresholding, which is the simplest
form of image segmentation.
In the next chapter, we will study and demonstrate video processing.

188
CHAPTER 10

Video Processing
In the previous chapter, we learned about using morphological
operations in the domain of image processing. We have also demonstrated
all those operations with libraries such as NumPy, SciPy, and matplotlib.
Till now, throughout this book, we have been discussing how to work
with static images. We know the means of acquiring static images and also
how to process them. This chapter is dedicated to processing streams of
continuous images, such as video and live webcam feed. We will also get
introduced to a new library, OpenCV. We will cover the following topics in
this chapter:

• Introduction to OpenCV library

• Converting the colorspace of image

• Separating color channels in a live webcam stream

• More operations on live webcam streams

By the end of this chapter, we will be very comfortable with processing

webcam feeds and videos with SciPy and NumPy.

Introduction to OpenCV Library

We have explored how many functions found in the SciPy and NumPy
libraries can be used to process images. We will now examine a very
limited part of another popular image processing library, OpenCV. We

© Ashwin Pajankar 2022 189

A. Pajankar, Raspberry Pi Image Processing Programming,
https://doi.org/10.1007/978-1-4842-8270-0_10
Chapter 10 Video Processing

are going to use it for reading live webcam stream and video files. For
processing the video and live webcam feed, we will still use SciPy.
OpenCV stands for open source computer vision. It is a library for
real-time computer vision. It includes a lot of functionality for image
processing and video procession. It also has implementations of many
computer vision algorithms. It is primarily written in C++. It has interfaces
(APIs; application programming interfaces) for languages such as C++,
Python, Java, MATLAB/OCTAVE, and JavaScript. We can explore this at the
webpage https://opencv.org/.
Let’s get started by installing the library on our Raspberry Pi with the
following command:

pip install opencv-python

For our further demonstrations, we will be needing a webcam. I

have been using a Logitech C922 Pro. The camera is quite costly, with a
price around 150 USD. However, I use it for multiple applications. If this
one is out of your budget, you can use a less expensive webcam for the
demonstrations. Most of the branded USB webcams will just do fine with
Raspberry Pi. You can also check the list of webcams compatible with RPi
at the webpage https://elinux.org/RPi_USB_Webcams. To get started,
connect the webcam to the Raspberry Pi using one of its four USB ports.
Now, have a look at the following code listing (Listing 10-1).

Listing 10-1.  prog00.py

import numpy as np
import cv2 as cv

cap = cv.VideoCapture(0)

cap.set(cv.CAP_PROP_FRAME_WIDTH, 256)
cap.set(cv.CAP_PROP_FRAME_HEIGHT, 144)

190
 Chapter 10 Video Processing

if not cap.isOpened():
    print("Cannot open camera...")
    exit()

while True:
    ret, frame = cap.read()

    if not ret:

        print("Can't receive frame... \nExiting ...")
        break

    cv.imshow('frame', frame)
    if cv.waitKey(1) == 27:
        break

cap.release()
cv.destroyAllWindows()

Execute the program. The indicator LED embedded in the webcam will
glow, and it will produce the following (Figure 10-1) output window on the
display,

Figure 10-1.  The output of the live webcam feed

191
Chapter 10 Video Processing

That’s me! We can stop the program by pressing the ESC key on the
keyboard. The window (Figure 10-1) will close, and the LED indicator in
the webcam will stop glowing. Now, it is time to understand the program
line-by-line. In the first two lines, we import the required libraries (which
includes OpenCV) and create aliases for our convenience. Then, the
following line creates an object corresponding to the webcam:

cap = cv.VideoCapture(0)

The argument passed is the index of the webcam. If only one webcam
is attached then the argument passed is always 0. The following two lines
define the resolution at which the webcam will operate:

cap.set(cv.CAP_PROP_FRAME_WIDTH, 256)
cap.set(cv.CAP_PROP_FRAME_HEIGHT, 144)

The function cap.set() sets the properties of the webcam connected

to the computer. There are plenty of properties. The following (https://
stackoverflow.com) question discusses the properties of webcams that
can be set with OpenCV in detail:

https://stackoverflow.com/questions/11420748/setting-camera-
parameters-in-opencv-python
Also, I wish to make readers aware that we will find the video stream
from the webcam faster and more responsive at lower resolutions. As the
processing power of the Raspberry Pi is limited, we will have more frames
per second (FPS) in the output window at lower resolutions. We can get a
list of all the standard resolutions at the following resources on the web:

https://en.wikipedia.org/wiki/List_of_common_resolutions
https://yusef.es/labs/resolutions.htm

The following lines of code terminate the execution of the program if

the webcam cannot be accessed:

192
 Chapter 10 Video Processing

if not cap.isOpened():
    print("Cannot open camera...")
    exit()

The while loop block runs as long as we do not hit the ESC key on the
keyboard.

while True:
    ret, frame = cap.read()

The function call cap.read() returns the status and the frame. The
frame is a numpy Ndarray object that has the current frame. We can
manipulate it as usual with NumPy and SciPy routines, just as we have
practiced in earlier chapters.
If the status of the webcam frame read operation is unsuccessful, then
we terminate the loop, as follows:

    if not ret:

        print("Can't receive frame... \nExiting ...")
        break

The following routine call shows the frame:

    cv.imshow('frame', frame)

The following code checks for the press of the ESC key on the
keyboard:

    if cv.waitKey(1) == 27:

        break

The following code releases the webcam by destroying the object:

cap.release()

The following code destroys all the current OpenCV windows:

cv.destroyAllWindows()

193
Chapter 10 Video Processing

Consider this program as the template for all the programs in the rest
of the chapter. We can modify the program to read and show a video file.
We have to modify the code where we address the webcam, as follows:

cap = cv.VideoCapture('test.mp4')

We are passing the name of the video file as the argument. We also
have to make modifications to the following line in the while loop block:

    if cv.waitKey(1) == 27:

The argument we are passing while using the webcam is 1. However,

to show the frames of the video file, we have to compute the value of this
argument we are passing to this function call. We first need to know the
frame rate of the video in terms of FPS and then divide 1000 by that rate.
Then, we have to pass the output of this computation as the argument to
the function call cv.waitKey(). Let us see a few examples with standard
frame rates. Usually, the cinematic frame rate is around 24 FPS (23.97,
actually). So, 1000/24 = 41.67, which is almost equal to 42. So, the function
call for 24 FPS will be cv.waitKey(42). For 25 FPS, it is cv.waitKey(40).
For 30 FPS, it is cv.waitKey(33), and so on. As an exercise, open a video
file by making the discussed modifications.

Converting the Colorspace of Image

Colorspace is a mathematical representation of color models. A color
model is a way to represent the colors in a reproducible way. The OpenCV
library reads an image or a live webcam feed or a video in the BGR
colorspace. We can convert an image from BGR to any colorspace of our
choice. We will frequently need to convert the live webcam feed from BGR
to grayscale throughout the chapter. Let’s see how to do that. Have a look at
Listing 10-2.

194
 Chapter 10 Video Processing

Listing 10-2.  prog01.py

import numpy as np
import cv2 as cv

cap = cv.VideoCapture(0)
cap.set(cv.CAP_PROP_FRAME_WIDTH, 256)
cap.set(cv.CAP_PROP_FRAME_HEIGHT, 144)

if not cap.isOpened():
    print("Cannot open camera...")
    exit()

while True:
    ret, frame = cap.read()

    if not ret:

        print("Can't receive frame... \nExiting ...")
        break

    gray = cv.cvtColor(frame,
                       cv.COLOR_BGR2GRAY)

    cv.imshow('frame', gray)
    if cv.waitKey(1) == 27:
        break

cap.release()
cv.destroyAllWindows()

In the following line, we are converting the color frame into a

greyscale frame,

    gray = cv.cvtColor(frame,
                       cv.COLOR_BGR2GRAY)

195
Chapter 10 Video Processing

The function cvtColor() is used to achieve the conversion between

colorspaces. We will also be using this function quite frequently
throughout the chapter as many image processing routines in SciPy need
grayscale images as the input. Run the program to see the output.

 eparating Color Channels in Live

S
Webcam Stream
We can also separate the color channels in the live webcam stream with
the code in Listing 10-3.

Listing 10-3.  prog02.py

import numpy as np
import cv2 as cv
from datetime import datetime

cap = cv.VideoCapture(0)

cap.set(cv.CAP_PROP_FRAME_WIDTH, 256)
cap.set(cv.CAP_PROP_FRAME_HEIGHT, 144)

if not cap.isOpened():
    print("Cannot open camera...")
    exit()

while True:
    ret, frame = cap.read()

    if not ret:

        print("Can't receive frame... \nExiting ...")
        break

    print(datetime.now())

196
 Chapter 10 Video Processing

    print(frame.dtype)
    print(frame.shape)
    print(frame.ndim)
    print(frame.size)

    b = frame[:, :, 0]
    g = frame[:, :, 1]
    r = frame[:, :, 2]

    cv.imshow('Red', r)
    cv.imshow('Green', g)
    cv.imshow('Blue', b)
    if cv.waitKey(1) == 27:
        break

cap.release()
cv.destroyAllWindows()

In this code listing (prog02.py), we can see the properties of each frame
as well as all the channels, as shown in the output (Figure 10-2).

Figure 10-2.  Seperation of color channels

Also, the console will show the output, like the one here, several times
per second (so, it will scroll really fast):

2022-01-30 13:00:30.427360

197
Chapter 10 Video Processing

uint8
(144, 176, 3)
3
76032

More Operations on Live Webcam Stream

Let us see more operations we can perform on a live webcam stream with
the SciPy library routines. We are going to cover a lot of examples. We
learned all these operations in earlier chapters, and we will now apply
them on a live webcam stream.

Geometric Transformation
Let’s rotate the webcam frame continuously. Listing 10-4 shows the code.

Listing 10-4.  prog03.py

import numpy as np
import cv2 as cv
import scipy.ndimage as ndi

angle = 0
cap = cv.VideoCapture(0)

cap.set(cv.CAP_PROP_FRAME_WIDTH, 256)
cap.set(cv.CAP_PROP_FRAME_HEIGHT, 144)

if not cap.isOpened():
    print("Cannot open camera...")
    exit()

while True:
    ret, frame = cap.read()

198
 Chapter 10 Video Processing

    if not ret:

        print("Can't receive frame... \nExiting ...")
        break

    angle = (angle + 1) % 360

    output = ndi.rotate(frame, angle,
                        reshape=False)

    cv.imshow('Rotated Frame', output)

    if cv.waitKey(1) == 27:
        break

cap.release()
cv.destroyAllWindows()

The code that manipulates the frame is the following one:

    angle = (angle + 1) % 360

    output = ndi.rotate(frame, angle,
                        reshape=False)

The logic is surprisingly plain. We just increment the rotation angle

with each frame, and when it reaches 360 degrees, we start over. The
following image (Figure 10-3) shows a screenshot of one of the frames.

Figure 10-3.  Geometric transformation on a live webcam feed

199
Chapter 10 Video Processing

Actually, the screenshot (Figure 10-3) really does not do any justice to

the real output. Run the program and see the output for yourself.

Convolution
Let’s define a kernel and apply it to the frames in the stream using the
convolution operation, as in Listing 10-5.

Listing 10-5.  prog04.py

import numpy as np
import cv2 as cv
import scipy.ndimage as ndi

cap = cv.VideoCapture(0)
n = 5
k = np.ones((n, n, n), np.uint8)/(n*n*n)

cap.set(cv.CAP_PROP_FRAME_WIDTH, 256)
cap.set(cv.CAP_PROP_FRAME_HEIGHT, 144)

if not cap.isOpened():
    print("Cannot open camera...")
    exit()

while True:
    ret, frame = cap.read()

    if not ret:

        print("Can't receive frame... \nExiting ...")
        break

    output = ndi.convolve(frame, k,
                          mode='nearest')

    cv.imshow('Convolved Frame', output)

200
 Chapter 10 Video Processing

    if cv.waitKey(1) == 27:

        break

cap.release()
cv.destroyAllWindows()

Since it is a blurring kernel, the output is a blurred feed from the

webcam, as shown in the screenshot (Figure 10-4).

Figure 10-4.  Blurred feed from the webcam

The OpenCV library also has the functionality to create a trackbar. Let’s
create a trackbar in such a way that it adjusts the size of the kernel matrix.
Listing 10-6 contains the code.

Listing 10-6.  prog05.py

import numpy as np
import cv2 as cv
import scipy.ndimage as ndi

def nothing(x):
    pass

WINDOW_NAME = "Live Convolution"

cv.namedWindow(WINDOW_NAME)
cv.createTrackbar('Kernel Size',

201
Chapter 10 Video Processing

                  WINDOW_NAME,
                  1, 5, nothing)
cap = cv.VideoCapture(0)
n = 1

cap.set(cv.CAP_PROP_FRAME_WIDTH, 256)
cap.set(cv.CAP_PROP_FRAME_HEIGHT, 144)

if not cap.isOpened():
    print("Cannot open camera...")
    exit()

while True:
    ret, frame = cap.read()

    if not ret:

        print("Can't receive frame... \nExiting ...")
        break

    n = cv.getTrackbarPos('Kernel Size',

                          WINDOW_NAME)
    k = np.ones((n, n, n), np.uint8)/(n*n*n)
    output = ndi.convolve(frame, k,
                          mode='nearest')

    cv.imshow(WINDOW_NAME, output)
    if cv.waitKey(1) == 27:
        break

cap.release()
cv.destroyAllWindows()

We are already familiar with most of the code. Let’s understand the
new parts. The following code defines an empty function:

def nothing(x):
    pass
202
 Chapter 10 Video Processing

The following code defines a named window:

WINDOW_NAME = "Live Convolution"

cv.namedWindow(WINDOW_NAME)

We need a named window to bind the newly created trackbar as

follows:

cv.createTrackbar('Kernel Size',
                  WINDOW_NAME,
                  1, 5, nothing)

This code creates a named trackbar that allows us to slide from 1 to 5.

It is attached to the earlier-created named window. The empty function
we defined earlier is associated with the trackbar and is called every time
we operate the trackbar. We get the current position of the slider with the
following function call in the while loop:

    n = cv.getTrackbarPos('Kernel Size', WINDOW_NAME)

We can now change the size of the kernel by sliding the tracker on the
trackbar. Figure 10-5 shows the output.

Figure 10-5.  Using trackbar

203
Chapter 10 Video Processing

We are going to use the code for the trackbar frequently throughout the
chapter.

Correlation
Let’s use the trackbar and apply a correlation on the live webcam feed, as
shown in Listing 10-7.

Listing 10-7.  prog06.py

import numpy as np
import cv2 as cv
import scipy.ndimage as ndi

def nothing(x):
    pass

WINDOW_NAME = "Live Correlation"

cv.namedWindow(WINDOW_NAME)
cv.createTrackbar('Kernel Size',
                  WINDOW_NAME,
                  1, 5, nothing)
cap = cv.VideoCapture(0)
n = 1

cap.set(cv.CAP_PROP_FRAME_WIDTH, 256)
cap.set(cv.CAP_PROP_FRAME_HEIGHT, 144)

if not cap.isOpened():
    print("Cannot open camera...")
    exit()

while True:
    ret, frame = cap.read()

204
 Chapter 10 Video Processing

    if not ret:

        print("Can't receive frame... \nExiting ...")
        break

    n = cv.getTrackbarPos('Kernel Size',

                          WINDOW_NAME)
    k = np.ones((n, n, n), np.uint8)/(n*n*n)
    output = ndi.correlate(frame, k,
                           mode='nearest')

    cv.imshow(WINDOW_NAME, output)
    if cv.waitKey(1) == 27:
        break

cap.release()
cv.destroyAllWindows()

Let’s see the output in Figure 10-6.

Figure 10-6.  Correlation with trackbar

205
Chapter 10 Video Processing

Note  We can read more about the concept of correlation at the

URL https://towardsdatascience.com/convolution-vs-
correlation-af868b6b4fb5.

Filtering
We have learned about many filters in earlier chapters. Let’s apply them on
the live webcam feed one by one. Listing 10-8 shows the Gaussian filter.

Listing 10-8.  prog07.py

import numpy as np
import cv2 as cv
import scipy.ndimage as ndi

def nothing(x):
    pass

WINDOW_NAME = "Live Gaussian"

cv.namedWindow(WINDOW_NAME)
cv.createTrackbar('Sigma Size',
                  WINDOW_NAME,
                  1, 15, nothing)
cap = cv.VideoCapture(0)

cap.set(cv.CAP_PROP_FRAME_WIDTH, 256)
cap.set(cv.CAP_PROP_FRAME_HEIGHT, 144)

if not cap.isOpened():
    print("Cannot open camera...")
    exit()

while True:

206
 Chapter 10 Video Processing

    ret, frame = cap.read()

    if not ret:

        print("Can't receive frame... \nExiting ...")
        break

    n = cv.getTrackbarPos('Sigma Size',

                          WINDOW_NAME)

    output = ndi.gaussian_filter(frame,
                                 sigma=n)

    cv.imshow(WINDOW_NAME, output)
    if cv.waitKey(1) == 27:
        break

cap.release()
cv.destroyAllWindows()

Let’s apply a uniform filter, as shown in Listing 10-9.

Listing 10-9.  prog08.py

import numpy as np
import cv2 as cv
import scipy.ndimage as ndi

def nothing(x):
    pass

WINDOW_NAME = "Live Uniform"

cv.namedWindow(WINDOW_NAME)
cv.createTrackbar('Filter Size',
                  WINDOW_NAME,
                  1, 15, nothing)
cap = cv.VideoCapture(0)

207
Chapter 10 Video Processing

cap.set(cv.CAP_PROP_FRAME_WIDTH, 256)
cap.set(cv.CAP_PROP_FRAME_HEIGHT, 144)

if not cap.isOpened():
    print("Cannot open camera...")
    exit()

while True:
    ret, frame = cap.read()

    if not ret:

        print("Can't receive frame... \nExiting ...")
        break

    n = cv.getTrackbarPos('Filter Size',

                          WINDOW_NAME)

    output = ndi.uniform_filter(frame,
                                size=n)

    cv.imshow(WINDOW_NAME, output)
    if cv.waitKey(1) == 27:
        break

cap.release()
cv.destroyAllWindows()

Let’s apply the percentile filter. The code in Listing 10-10 defines two
trackbars for adjusting the arguments passed to the filter.

Listing 10-10.  prog09.py

import numpy as np
import cv2 as cv
import scipy.ndimage as ndi

def nothing(x):
    pass
208
 Chapter 10 Video Processing

WINDOW_NAME = "Live Percentile"

cv.namedWindow(WINDOW_NAME)
cv.createTrackbar('Size',
                  WINDOW_NAME,
                  1, 5, nothing)
cv.createTrackbar('Percentile',
                  WINDOW_NAME,
                  1, 75, nothing)
cap = cv.VideoCapture(0)

cap.set(cv.CAP_PROP_FRAME_WIDTH, 256)
cap.set(cv.CAP_PROP_FRAME_HEIGHT, 144)

if not cap.isOpened():
    print("Cannot open camera...")
    exit()

while True:
    ret, frame = cap.read()

    if not ret:

        print("Can't receive frame... \nExiting ...")
        break

    s = cv.getTrackbarPos('Size',
                          WINDOW_NAME)

    p = cv.getTrackbarPos('Percentile',
                          WINDOW_NAME)

    output = ndi.percentile_filter(frame, size=s,

                                   percentile=p)

    cv.imshow(WINDOW_NAME, output)
    if cv.waitKey(1) == 27:
        break

209
Chapter 10 Video Processing

cap.release()
cv.destroyAllWindows()

The output is as shown in Figure 10-7.

Figure 10-7.  Percentile filter

Let’s implement a minimum filter as shown in Listing 10-11.

Listing 10-11.  prog10.py

import numpy as np
import cv2 as cv
import scipy.ndimage as ndi

def nothing(x):
    pass

WINDOW_NAME = "Live Minimum"

cv.namedWindow(WINDOW_NAME)
cv.createTrackbar('Filter Size',
                  WINDOW_NAME,
                  1, 55, nothing)

210
 Chapter 10 Video Processing

cap = cv.VideoCapture(0)

cap.set(cv.CAP_PROP_FRAME_WIDTH, 256)
cap.set(cv.CAP_PROP_FRAME_HEIGHT, 144)

if not cap.isOpened():
    print("Cannot open camera...")
    exit()

while True:
    ret, frame = cap.read()

    if not ret:

        print("Can't receive frame... \nExiting ...")
        break

    n = cv.getTrackbarPos('Filter Size',

                          WINDOW_NAME)

    output = ndi.minimum_filter(frame,
                                size=n)

    cv.imshow(WINDOW_NAME, output)
    if cv.waitKey(1) == 27:
        break

cap.release()
cv.destroyAllWindows()

Let’s apply the maximum filter, as in Listing 10-12.

Listing 10-12.  prog11.py

import numpy as np
import cv2 as cv
import scipy.ndimage as ndi

211
Chapter 10 Video Processing

def nothing(x):
    pass

WINDOW_NAME = "Live Maximum"

cv.namedWindow(WINDOW_NAME)
cv.createTrackbar('Filter Size',
                  WINDOW_NAME,
                  1, 25, nothing)
cap = cv.VideoCapture(0)

cap.set(cv.CAP_PROP_FRAME_WIDTH, 256)
cap.set(cv.CAP_PROP_FRAME_HEIGHT, 144)

if not cap.isOpened():
    print("Cannot open camera...")
    exit()

while True:
    ret, frame = cap.read()

    if not ret:

        print("Can't receive frame... \nExiting ...")
        break

    n = cv.getTrackbarPos('Filter Size',

                          WINDOW_NAME)

    output = ndi.maximum_filter(frame,
                                size=n)

    cv.imshow(WINDOW_NAME, output)
    if cv.waitKey(1) == 27:
        break

cap.release()
cv.destroyAllWindows()

212
 Chapter 10 Video Processing

We can also apply Prewitt and Laplace filters as shown in Listing 10-13.

Listing 10-13.  prog12.py

import numpy as np
import cv2 as cv
import scipy.ndimage as ndi

cap = cv.VideoCapture(0)

cap.set(cv.CAP_PROP_FRAME_WIDTH, 256)
cap.set(cv.CAP_PROP_FRAME_HEIGHT, 144)

if not cap.isOpened():
    print("Cannot open camera...")
    exit()

while True:
    ret, frame = cap.read()

    if not ret:

        print("Can't receive frame... \nExiting ...")
        break

    o1 = ndi.prewitt(frame)
    o2 = ndi.laplace(frame)

    cv.imshow("Prewitt", o1)
    cv.imshow("Laplace", o2)
    if cv.waitKey(1) == 27:
        break

cap.release()
cv.destroyAllWindows()

213
Chapter 10 Video Processing

Note  We can read more about both filters at the following URLs:
https://www.geeksforgeeks.org/matlab-image-edge-
detection-using-prewitt-operator-from-scratch/
https://www.l3harrisgeospatial.com/docs/
laplacianfilters.html

Let’s apply the Gaussian gradient magnitude filter as in Listing 10-14.

Listing 10-14.  prog13.py

import numpy as np
import cv2 as cv
import scipy.ndimage as ndi

def nothing(x):
    pass

WINDOW_NAME = "Gaussian Gradient Magnitude"

cv.namedWindow(WINDOW_NAME)
cv.createTrackbar('Sigma',
                  WINDOW_NAME,
                  1, 10, nothing)
cap = cv.VideoCapture(0)

cap.set(cv.CAP_PROP_FRAME_WIDTH, 256)
cap.set(cv.CAP_PROP_FRAME_HEIGHT, 144)

if not cap.isOpened():
    print("Cannot open camera...")
    exit()

while True:
    ret, frame = cap.read()

214
 Chapter 10 Video Processing

    if not ret:

        print("Can't receive frame... \nExiting ...")
        break

    s = cv.getTrackbarPos('Sigma',
                          WINDOW_NAME)

    gray = cv.cvtColor(frame, cv.COLOR_BGR2GRAY)

    output = ndi.gaussian_gradient_magnitude(gray,
                                             sigma=s)

    cv.imshow(WINDOW_NAME, output)
    if cv.waitKey(1) == 27:
        break

cap.release()
cv.destroyAllWindows()

Let’s apply the Gaussian Laplace filter as in Listing 10-15.

Listing 10-15.  prog14.py

import numpy as np
import cv2 as cv
import scipy.ndimage as ndi

def nothing(x):
    pass

WINDOW_NAME = "Gaussian Laplace"

cv.namedWindow(WINDOW_NAME)
cv.createTrackbar('Sigma',
                  WINDOW_NAME,
                  1, 10, nothing)
cap = cv.VideoCapture(0)

215
Chapter 10 Video Processing

cap.set(cv.CAP_PROP_FRAME_WIDTH, 256)
cap.set(cv.CAP_PROP_FRAME_HEIGHT, 144)

if not cap.isOpened():
    print("Cannot open camera...")
    exit()

while True:
    ret, frame = cap.read()

    if not ret:

        print("Can't receive frame... \nExiting ...")
        break

    s = cv.getTrackbarPos('Sigma',
                          WINDOW_NAME)

    gray = cv.cvtColor(frame, cv.COLOR_BGR2GRAY)

    output = ndi.gaussian_laplace(gray,
                                  sigma=s)

    cv.imshow(WINDOW_NAME, output)
    if cv.waitKey(1) == 27:
        break

cap.release()
cv.destroyAllWindows()

Morphological Operations
We can also apply morphological operations on live video, as shown in
Listing 10-16.

216
 Chapter 10 Video Processing

Listing 10-16.  prog15.py

import numpy as np
import cv2 as cv
import scipy.ndimage as ndi

def nothing(x):
    pass

WINDOW_NAME = "Distance Transform"

cv.namedWindow(WINDOW_NAME)

cv.createTrackbar('Sigma',
                  WINDOW_NAME,
                  1, 10, nothing)

cap = cv.VideoCapture(0)

cap.set(cv.CAP_PROP_FRAME_WIDTH, 256)
cap.set(cv.CAP_PROP_FRAME_HEIGHT, 144)

if not cap.isOpened():
    print("Cannot open camera...")
    exit()

while True:
    ret, frame = cap.read()

    if not ret:

        print("Can't receive frame... \nExiting ...")
        break

    grey = cv.cvtColor(frame, cv.COLOR_BGR2GRAY)

    n1 = cv.getTrackbarPos('Sigma',
                          WINDOW_NAME)

217
Chapter 10 Video Processing

    output = ndi.grey_dilation(grey, size=(7, 7),

                               structure=np.ones((n1, n1)))

    cv.imshow(WINDOW_NAME, output)
    if cv.waitKey(1) == 27:
        break

cap.release()
cv.destroyAllWindows()

This code applies the grayscale dilation to a live webcam stream, as

shown in Figure 10-8.

Figure 10-8.  Grayscale dilation

We can similarly apply erosion to a live feed, as shown in Listing 10-17.

Listing 10-17.  prog16.py

import numpy as np
import cv2 as cv
import scipy.ndimage as ndi

def nothing(x):
    pass

218
 Chapter 10 Video Processing

WINDOW_NAME = "Distance Transform"

cv.namedWindow(WINDOW_NAME)

cv.createTrackbar('Sigma',
                  WINDOW_NAME,
                  1, 10, nothing)

cap = cv.VideoCapture(0)

cap.set(cv.CAP_PROP_FRAME_WIDTH, 256)
cap.set(cv.CAP_PROP_FRAME_HEIGHT, 144)

if not cap.isOpened():
    print("Cannot open camera...")
    exit()

while True:
    ret, frame = cap.read()

    if not ret:

        print("Can't receive frame... \nExiting ...")
        break

    grey = cv.cvtColor(frame, cv.COLOR_BGR2GRAY)

    n1 = cv.getTrackbarPos('Sigma',
                          WINDOW_NAME)

    output = ndi.grey_erosion(grey, size=(7, 7),

                              structure=np.ones((n1, n1)))

    cv.imshow(WINDOW_NAME, output)
    if cv.waitKey(1) == 27:
        break

219
Chapter 10 Video Processing

cap.release()
cv.destroyAllWindows()

The output is as seen in Figure 10-9.

Figure 10-9.  Grayscale erosion

Summary
In this chapter, we learned how to apply the SciPy routines to a live video
feed from a webcam and on video files. We are now comfortable with
processing videos (both live and stored files).

Conclusion
In this book, we covered various Python libraries meant for image
processing. We also used the Raspberry Pi 4 single-board computer
with the Raspberry Pi OS Debian Buster as the preferred platform for
demonstrating all the Python 3 code. All the code examples can be run on
any other operating system as well, such as Windows, macOS, other Linux
distributions, Unix distributions, and BSD (such as FreeBSD).

220
 Chapter 10 Video Processing

The world of image processing and computer vision is very vast. If we

consider it as an ocean, then we are still on the shores of it. Equipped with
basic knowledge, we can definitely venture ourselves into deeper waters
and gain understanding of computer vision in greater depth. Armed with
the advanced knowledge, we can program various real-life examples for
computer vision.

221
Appendix
This section examines topics that will help us to work with image
processing and Raspberry Pi SBC in a more efficient way. While I could not
find a suitable place for these topics in any of the chapters, I personally feel
that they are quite useful in the learning process. This appendix section
covers the following topics:

• pgmagick image-processing library

• Connecting a display

• Connecting to ethernet/wired network

• Remote desktop with Virtual Network

Computing (VNC)

After reviewing the appendix, you should be comfortable with these

tips and tricks for Raspberry Pi.

pgmagick Image Processing

Here we will work with a simple image-processing library known as
pgmagick. It is a Python wrapper around the image-manipulation
program ImageMagik. One can read more about the ImageMagik program
at its homepage: https://imagemagick.org/index.php.
Let’s install everything by running the following commands in sequence:

sudo apt-get install imagemagick libgraphicsmagick++1-dev -y

sudo apt-get install g++ libboost-python-dev -y
sudo pip3 install pgmagick

© Ashwin Pajankar 2022 223

A. Pajankar, Raspberry Pi Image Processing Programming,
https://doi.org/10.1007/978-1-4842-8270-0
Appendix

It installs everything required to write a Python program using the

pgmagick library. Listing A-1 creates a simple background with green color
and saves that as an image in the current directory.

Listing A-1.  prog00.py

from pgmagick.api import Image

img = Image((300, 200), 'Green')
img.write('bg-green.png')

Let’s create a transparent background as in Listing A-2.

Listing A-2.  prog01.py

from pgmagick.api import Image

img = Image((300, 200), 'transparent')
img.write('transparent.png')

Let’s create a gradient image as shown in Listing A-3.

Listing A-3.  prog02.py

from pgmagick.api import Image

img = Image((300, 200),
            'gradient:#00ff00-#0000ff')
img.write('gradient.png')

We can add some text to an image with the code shown in Listing A-4.

Listing A-4.  prog03.py

from pgmagick.api import Image

img = Image((300, 200))
img.annotate('Hello World', angle=60)
img.write('helloworld60.png')

224
 Appendix

We can also retrieve the properties of an image as shown in Listing A-5.

Listing A-5.  prog04.py

from pgmagick.api import Image

img = Image((300, 200))
print (img.width, img.height)

These are the basics of the library. Please run all the preceding
programs and see the output. One can explore this library further at
https://pythonhosted.org/pgmagick/.

Connecting a Display
Up until now, we have been accessing the Raspberry Pi in headless
mode (connecting using SSH and X-11 forwarding using Wi-Fi). We can
also connect a visual display (an HDMI/VGA monitor) by choosing the
appropriate connector. If we are opting for HDMI with RPi 4B, then we
need a micro-HDMI to HDMI adapter (check online marketplaces for
this). If we are opting for a VGA monitor, we need a micro-HDMI to VGA
adapter. If we check online marketplaces, we can find a micro-HDMI to
HDMI and VGA combined adapter that takes care of both cases.

Using a VGA Display

Additionally, if we wish to use a VGA display, then we must make changes
to the /boot/config.txt file before booting up the first time. This step is
needed only for a VGA monitor. If you are using an HDMI monitor, then
skip this part.
Insert the microSD card into the card reader and browse it in Windows
Explorer (or the file explorer of your operating system). In Windows
Explorer, it will be represented as a removable media drive labeled as boot.

225
Appendix

Open the config.txt file and make the following changes to it:

• Change #disable_overscan=1 to disable_overscan=1

• Change #hdmi_force_hotplug=1 to hdmi_force_

hotplug=1

• Change #hdmi_group=1 to hdmi_group=2

• Change #hdmi_mode=1 to hdmi_mode=16

• Change #hdmi_drive=2 to hdmi_drive=2

• Change #config_hdmi_boost=4 to config_

hdmi_boost=4

Save the file after making these changes. The microSD card is now
ready for the Pi with a VGA monitor.

Booting Up After Connecting a Display

As we have seen, we can connect the Pi to the monitor of our choice and
boot up. After booting up for the very first time, we will be greeted by a
setup or configuration wizard, as shown in Figure A-1. It is a one-time
process only, so go ahead.

226
 Appendix

Figure A-1.  Configuration wizard

227
Appendix

Click on the Next button, and you will be shown the following window
(Figure A-2).

Figure A-2.  Localization settings

228
 Appendix

Choose the most appropriate options for your region and click Next
(you can also navigate to the previous options by clicking on Back). You
will see the following (Figure A-3) window.

Figure A-3.  Changing the default password

229
Appendix

Change the default password. It is strongly recommended, as anyone

who hacks into your Wi-Fi can remotely log in to your Pi using the default
password and wreak havoc. Press the Next button, and it will show the
following (Figure A-4) window.

Figure A-4.  Set up screen

230
 Appendix

Check the box if the taskbar (in the top area of the screen) does
not fit the screen, and then click on Next. It shows the following
(Figure A-5) window.

Figure A-5.  Set up the Wi-Fi

231
Appendix

We can set up the Wi-Fi here (if not set up already; if it is, click on the
Skip button), and it shows the update window (Figure A-6).

Figure A-6.  Software update

232
 Appendix

We can click on the Next button, and it will update the software (we
can also skip this, but I recommend updating the software). Once done, it
will show a success message (Figure A-7).

Figure A-7.  Success!

233
Appendix

Here, click on the Restart button. After reboot, if you have skipped
changing the password, it will show the following warning (Figure A-8).

Figure A-8.  Warning

We will continue seeing this warning after booting up as long as the

password is the default one. Once we change the password, it will stop
appearing after bootup.
This finishes setup.

Connecting to Ethernet/ Wired Network

Most of us will use the Pi boards in an organization (workplace or
academic institution) or at home. These places will always have wired
LANs that are connected to the internet. We can connect the Pi to a
LAN using the RJ45 connector. Pi boards have an RJ45 port for wired
networking. Almost all modern routers come with DHCP (Dynamic Host
Configuration Protocol), and an IP address will automatically be assigned
to the Pi once connected to a wired network.

234
 Appendix

Remote Desktop with VNC

We can access the desktop remotely using the VNC software. We have to
enable it in the Pi before using it. We saw the raspi-config tool in the very
first chapter. We can enable VNC using that (in the command line). We can
also find a GUI utility for the same task in the desktop environment. Click
the Raspberry symbol in the top left. It is a Windows-like menu. In the
Preferences section, we can find the option Raspberry Pi Configuration.
Click on that, and it will open a new window. There, click on the third tab,
labeled Interfaces. It shows the following (Figure A-9).

Figure A-9.  Interfaces

Here, make sure that at least the options SSH and VNC are enabled.
I need other options for my development and programming lessons, so I
enabled them all.

235
Appendix

After this, download a VNC viewer client from any of the

following URLs:
https://www.realvnc.com/en/connect/download/viewer/
https://www.tightvnc.com/download.php
https://sourceforge.net/projects/tigervnc/files/
stable/1.12.0/
RealVNC is one of the most popular VNC clients and the one I use.
I will explain the setup process for this. If you wish to use the other
clients, please check the relevant documentation and tutorials. We first
have to configure a connection by providing the IP address, as follows
(Figure A-10).

236
 Appendix

Figure A-10.  Configure the remote connection

We also have to authenticate. I prefer to save the password so I won’t

have to enter it every time I try to connect (Figure A-11).

237
Appendix

Figure A-11.  Enter and save the password

Finally, if you are connecting for the very first time, it will show a
warning message. Click on the Always button, and it will launch the
remote desktop (Figure A-12).

238
 Appendix

Figure A-12.  Raspberry Pi OS with the remote desktop using

RealVNC

Figure A-12 shows a remote desktop with a terminal client opened. In

the top section, we can see the options for the VNC software. Explore them
for better understanding.

239
Index
A E
Analog image processing, 47 Equalization, 110, 111
Ethernet/ Wired Network, 234

B F
Binary thresholding, 182, 183
Filters, 145
Box blur, 105
convolution, 146, 148
correlation, 149
C fourier filters, 165
high-pass filters
Color quantization, 108, 109
Prewitt filter, 162
Colorspace conversion, 71, 72
Sobel filters, 164
Convolution, 95–97, 146–148,
kernel, 146
150, 161
low-pass filters, 150
Custom Gaussian filter, 93
blurring kernels, 150
noise reduction, 156, 158,
D 159, 161
fswebcam, 56, 57
Digital image processing (DIP),
31, 46–48
image processing, 47 G
Python 3 for, 61–67 Gaussian filter, 93, 150, 151, 153,
Raspberry Pi and Python for, 49 156–158, 206
signal processing, 46, 47 Geany, 43, 44, 50
Digital unsharp, 97–99 Grayscale morphological
Distance transform, 169, 170, operations, 178, 180, 181
180, 188 guvcview, 55, 56

© Ashwin Pajankar 2022 241

A. Pajankar, Raspberry Pi Image Processing Programming,
https://doi.org/10.1007/978-1-4842-8270-0
index

H ImageOps, 67, 69, 89, 90

Image processing programming
Histogram equalization, 111, 112
analog image processing, 47
Histograms, 90, 91, 110, 111
fswebcam, 56, 57
guvcview, 55, 56
I image sources, 51–53
Image Pi Camera Module, 57–60
blending, 72–74 raspistill, 61
channels, 70 Webcam, 54, 55
color quantization, 108, 109 Image sources, 51–53
histogram equalization, Integrated development
111, 112 environment (IDEs)
histograms and equalization, Geany, 43, 44
110, 111 IDLE, 40–42
ImageChops, 69 Thonny IDE, 45
ImageChops module, 85–88
ImageEnhance module,
J
91, 105–108
Julia set, 81–83
ImageFilter module, 91–105
Image.fromarray() method, 126
Image module, 69 K
particular pixel, knowing Kernel, 95–97, 143
value of, 79
colorspace conversion, 71, 72
copying and saving, 79 L
crop and paste Live webcam stream
operations, 77, 78 convolution operation, 200
image blending, 72–74 correlation, 204
image channels, 70 filtering, 206
Julia set, 81–83 Gaussian gradient
Mandelbrot set, 79–81 magnitude filter, 214
noise and gradients, 84 Gaussian Laplace filter, 215
resizing, 75, 76 maximum filter, 211
rotating, 76, 77 percentile filter, 208

242
 index

Prewitt and Laplace N, O

filters, 213
Noise and gradients, 84
uniform filter, 207
geometric transformation, 198
grayscale dilation, 218, 220 P, Q
morphological pgmagick Image
operations, 216 Processing, 223–225
Pi camera module, 49, 51, 57–60
Power supply, 11, 20
M Python, 3, 31, 37
Mandelbrot set, 79–81 digital image processing, 61–67
Matplotlib, 118 DIP, 49
image channels, 123 Python Imaging Library (PIL), 61
np.dstack() method, 125 Python programming language
pyplot module, 119 features of, 34–37
subplot() method, 122 Geany, 43, 44
Max filter, 99, 102, 103 history of, 32, 33
Measurements, 139 IDLE, 40–42
center of mass, 142 integrated development
histogram, 140 environment, 40
statistical properties, 141 interactive mode, 38, 39
Median filter, 99, 101, 158 Python 3, 37
microSD card reader, 10, 11, 20 Raspberry Pi OS, Python 2 and
Min filter, 101, 102 Python 3 on, 38
Mode filter, 103, 104 script mode, 39
Morphological gradient, 186, 187 Thonny IDE, 45
Morphological operations, 171
binary image, 173
binary propagation, 177 R
dark pixels, 174 Rapid Prototyping, 36
grey_dilation() method, 178, Raspberry Pi, 5–9
180, 181 accessing remotely, 20–25
noise removal, 174 booting up, 20
structuring element, 172, 176 config.txt File, 27, 28

243
index

Raspberry Pi (Cont.) Signal processing, digital image

configuration, 25, 26 processing, 46, 47
DIP, 49 Single-board computers
firmware updation, 28 (SBCs), 1, 2
microSD card, popular SBC families, 5
preparation, 12–20 Raspberry Pi, 5–29
Raspberry Pi OS, updating and and regular computers, 3
upgrading, 28, 29 SoCs, 3, 4
setup, 10, 11 System on Chips (SoCs), 3
shutting down and history of, 4
restarting RPi, 29 vs. regular CPU, 4
Raspberry Pi OS, Python 2 and
Python 3 on, 38
raspistill, 61
T, U
Remote Desktop, VNC Thonny IDE, 45
interfaces, 235 Thresholding and segmentation,
preferences section, 235 182, 184
Raspberry Pi OS, 239 TKinter, 50, 51, 63, 64, 69,
RealVNC, 236 72, 73, 95
Robust, 36 Transformations, 129
affine transformation, 136
geometric transformation, 134
S rotate() function, 132
Scientific Python, 113–116, 127 shift() method, 130
Scientific Python ecosystem, 114 spline transformation, 137
Core components, 115 zoom() method, 131
dtype attribute, 117
grayscale image, 118
NumPy homepage, 116 V
numpy.ndarray object, 117 VGA display, 225
NumPy’s N-dimensional booting up, 226
array, 116 change default password, 229
scientific computing, 115 config.txt file, 226
scipy.misc, store, 115 configuration wizard, 227

244
 index

localization settings, 228 Logitech C922 Pro, 190

Restart button, 234 OpenCV, 190
software update, 232
Video processing, 189 W, X, Y
color channels, live webcam
Webcam, 54, 55
stream, 196
White and black tophat, 184, 185
Colorspace, 194
Wi-Fi connection, 10
FPS, 192
LED indicator, 192
live webcam stream (see Live Z
webcam stream) Zenmap, 22

245