Unit 5: Computer Vision

Computer Vision (CV) is an important domain of Artificial Intelligence that enables machines to understand and interpret visual information such as images and videos.

In this unit, you will learn how computers process images, identify objects, recognise patterns and use Artificial Intelligence to make decisions from visual data.

Topics Covered in This Unit

1. 5.1 Introduction to Computer Vision
   • CV open-source tool
   • A quick overview of Computer Vision
   • Computer Vision and Artificial Intelligence
   • Computer Vision vs Image Processing
2. 5.2 Applications of Computer Vision
   • Facial Recognition
   • Face Filters
   • Google Search by Image
   • Computer Vision in Retail
   • Self-Driving Cars
   • Medical Imaging
   • Google Translate App
3. 5.3 Computer Vision Tasks
   • Classification
   • Classification + Localisation
   • Object Detection
   • Image Segmentation
   • Basics of Images and Pixels
   • Resolution and Pixel Value
   • Grayscale and RGB Images
4. 5.4 No-Code AI Tools
   • Introduction to Lobe
   • Teachable Machine
   • Smart Sorter Activity
   • Orange Data Mining Tool
   • Use Case Walkthrough
   • Steps to Project Development
5. 5.5 Image Features
   • Introduction to Image Features
   • Examples
   • Conclusion
6. 5.6 Convolution
   • Convolution
   • What is a Kernel?
7. 5.7 Convolutional Neural Network
   • Introduction
   • Convolution Layer
   • Rectified Linear Unit
   • Pooling Layer
   • Fully Connected Layer
8. 5.8 Python Libraries in Computer Vision
   • TensorFlow
   • Keras
   • OpenCV
   • Applications of OpenCV

5.1 Introduction to Computer Vision

What is Computer Vision?

Computer Vision (CV) is a field of Artificial Intelligence that enables computers to identify, process and understand information present in images and videos.

In simple words, Computer Vision enables a machine to "see" and understand the visual world.

Human Vision vs Computer Vision

Human	Computer
Eyes capture visual information.	Camera or sensor captures visual information.
Brain processes the information.	Computer processes pixels and data.
Brain identifies objects.	AI model identifies objects.
Human understands the scene.	Computer Vision interprets the visual data.

CV Open-Source Tool

One of the most popular open-source tools used for Computer Vision is OpenCV.

OpenCV stands for Open Source Computer Vision Library. It is a library containing functions and tools for processing images and videos and developing Computer Vision applications.

OpenCV can be used for tasks such as:

• Reading and displaying images
• Processing videos
• Working with webcams
• Face detection
• Object detection
• Image processing
• Real-time Computer Vision applications

Computer Vision and Artificial Intelligence

Computer Vision is a domain of Artificial Intelligence that focuses on understanding visual information.

Artificial Intelligence enables machines to perform tasks that normally require human intelligence. Computer Vision provides AI systems with the ability to analyse images and videos.

Computer Vision vs Image Processing

Image Processing mainly deals with manipulating or enhancing an image, whereas Computer Vision goes further by extracting meaningful information from visual data.

Computer Vision	Image Processing
Focuses on understanding visual information.	Focuses mainly on manipulating or enhancing images.
Can identify objects or people.	Can change brightness, contrast or size.
Example: Identifying a student from a photograph.	Example: Cropping or resizing a photograph.
Can make predictions from visual data.	Usually prepares or modifies the visual data.

Quick Examples

• Increasing brightness of a photograph → Image Processing
• Detecting a person's face → Computer Vision
• Cropping an image → Image Processing
• Identifying a car in an image → Computer Vision

Case Study: Smart Safety Car

A self-driving car uses a camera at night. The system successfully increases the brightness and contrast of the captured image but fails to identify a pedestrian.

The brightness and contrast adjustment is an example of Image Processing, while identifying the pedestrian requires Computer Vision.

5.2 Applications of Computer Vision

Computer Vision is used in many areas of our daily lives. It helps computers analyse images and videos and perform tasks that previously required human visual intelligence.

1. Facial Recognition

Facial Recognition is a Computer Vision application that identifies or verifies a person using facial characteristics.

Examples include:

• Smartphone face unlock
• Security systems
• Automated attendance systems
• Identity verification

2. Face Filters

Face filters use Computer Vision to identify facial landmarks such as the eyes, nose, mouth and face boundary.

Digital objects such as glasses, hats, masks or animal ears can then be placed over the face.

Example: Social media applications use Computer Vision to apply face filters in real time.

3. Google Search by Image

Search by Image allows users to search for information using an image instead of typing a text query.

Computer Vision analyses visual characteristics of the image and compares them with available visual information.

4. Computer Vision in Retail

Retail businesses use Computer Vision to analyse customers, products and shelves.

• Monitoring customer movement
• Checking product availability
• Inventory management
• Analysing shelf arrangements
• Improving shopping experiences

5. Self-Driving Cars

Self-driving cars use cameras and other sensors to understand their surroundings.

Computer Vision can help detect:

• Pedestrians
• Other vehicles
• Traffic lights
• Road signs
• Road obstacles

6. Medical Imaging

Computer Vision can analyse medical images such as X-rays, CT scans and MRI images to assist medical professionals.

Such systems can help identify patterns and abnormalities in medical images and support doctors in analysis.

7. Google Translate App

Google Translate can use a smartphone camera to recognise text contained in an image and translate it into another language.

Application Summary

Application	Use of Computer Vision
Facial Recognition	Identifying or verifying people
Face Filters	Detecting facial landmarks
Google Search by Image	Searching using visual information
Retail	Product and customer analysis
Self-Driving Cars	Detecting vehicles, pedestrians and road objects
Medical Imaging	Analysing medical images
Google Translate	Recognising and translating text from images

Case Study: Smart Library

A school library installs a camera at the book-return counter. The system identifies whether a book belongs to Science, History or Fiction and places a box around the book.

The system is using Computer Vision for classification and localisation/object detection.

5.3 Computer Vision Tasks

Computer Vision performs different tasks depending upon what information needs to be extracted from an image.

1. Classification

Classification assigns an image to a particular class or category.

It answers the question: "What is this image?"

Example: An image is classified as "Cat".

Other examples:

• Rose / Lily / Sunflower
• Healthy / Diseased
• Cat / Dog
• Fruit / Junk Food

2. Classification + Localisation

Classification + Localisation identifies what an object is and where it is located in an image.

It is generally used when there is a single object of interest.

Example: Identify a cat and draw a bounding box around it.

3. Object Detection

Object Detection identifies and locates multiple objects in an image or video.

Each detected object can have its own label and bounding box.

Example: A self-driving car may detect a pedestrian, two cars and a traffic light in the same image.

4. Image Segmentation

Image Segmentation divides an image into meaningful regions or identifies the pixels belonging to particular objects.

Instead of simply drawing a rectangular box around an object, segmentation can identify the exact shape of the object at the pixel level.

Computer Vision Tasks – Quick Comparison

Task	What does it answer?	Example
Classification	What is the object/image?	Cat
Classification + Localisation	What is it and where is it?	Cat + one bounding box
Object Detection	What objects are present and where?	Car + person + traffic light
Image Segmentation	Which pixels belong to each object?	Exact shape of a person

Basics of Images and Pixels

What is a Pixel?

The word Pixel is derived from Picture Element.

A pixel is the smallest unit of information in a digital image. Digital images are made up of many pixels arranged in a two-dimensional grid.

Resolution

Resolution refers to the number of pixels present in an image, generally represented by its width and height.

For example, a resolution of 1920 × 1080 means:

• Width = 1920 pixels
• Height = 1080 pixels

Total pixels: 1920 × 1080 = 2,073,600 pixels

Megapixel

1 Megapixel (MP) = 1 million pixels.

Therefore, a 12 MP camera can capture approximately 12 million pixels in an image.

Pixel Value

A pixel has a numerical value representing its brightness or colour.

In a standard 8-bit image, pixel values range from 0 to 255.

Pixel Value	Meaning in Grayscale
0	Black
128	Medium Gray
255	White

Grayscale Images

A grayscale image contains shades of gray rather than colours.

Each pixel generally has one value between 0 and 255.

• 0 → Black
• 255 → White
• Values between 0 and 255 → Different shades of gray

RGB Images

A coloured image can be represented using three colour channels: Red, Green and Blue (RGB).

Each pixel contains three values: one for Red, one for Green and one for Blue.

R	G	B	Result
0	0	0	Black
255	255	255	White
255	0	0	Red
0	255	0	Green
0	0	255	Blue

Case Study: Smart Refrigerator

A smart refrigerator has a camera that identifies Milk, Apple and Eggs and places a box around each item.

This is an example of Object Detection because multiple objects are identified and located.

If the system had to identify the exact pixels belonging to the liquid inside a bottle, it would require Image Segmentation.

5.4 No-Code AI Tools in Computer Vision

Traditional AI development may require programming knowledge. No-Code AI tools provide visual interfaces that allow users to create AI models with little or no programming.

Introduction to Lobe

Lobe is a no-code/low-code machine learning tool that makes it easier to train image classification models.

Users can provide labelled images and train a model without manually writing the machine learning algorithm.

Typical process:

1. Collect images.
2. Assign labels to images.
3. Train the model.
4. Test the model.
5. Use or export the trained model.

Teachable Machine

Teachable Machine is a web-based tool developed by Google that allows beginners to train simple machine learning models.

It can be used for:

• Image recognition
• Sound recognition
• Pose recognition

It is particularly useful for students because models can be trained using a simple visual interface.

Smart Sorter Activity

The Smart Sorter activity demonstrates how Computer Vision can be used to automatically classify different types of waste.

Suppose a smart waste sorter has three categories:

• Bottles
• Cans
• Paper

The project can follow these steps:

1. Collect images of bottles, cans and paper.
2. Label the images according to their categories.
3. Train the AI model.
4. Test the model with new images.
5. Improve the training data if incorrect predictions occur.

Orange Data Mining Tool

Orange Data Mining is an open-source tool that allows users to build data analysis and machine learning workflows using a visual interface.

Orange uses Widgets as building blocks. Widgets can be connected on a visual canvas to create a workflow.

Widget	Purpose
File	Load or import data
Data Table	View data in rows and columns
Scatter Plot	Visualise relationships between variables
Test & Score	Evaluate machine learning models

Use Case Walkthrough: Coral Bleaching

Coral reefs are important marine ecosystems. Environmental stress can cause coral bleaching.

Computer Vision can help researchers classify coral images as Healthy or Bleached.

A no-code tool can be used to train a model using labelled images. New images can then be tested to determine whether the coral appears healthy or bleached.

Steps to Project Development

A Computer Vision project using a no-code tool can follow the AI Project Cycle.

1. Data Acquisition: Collect or import suitable images.
2. Data Exploration: Examine the images and check whether the data is useful and properly labelled.
3. Modelling: Train an appropriate machine learning model.
4. Evaluation: Test the model and determine how well it performs.

Case Study: Cafeteria Fruit Scanner

A school cafeteria uses Lobe to classify food as Fruit or Junk Food. The system frequently classifies oranges as cookies because both can appear round and yellowish.

The problem may be caused by insufficient or unsuitable training data. The model needs more representative images showing oranges and cookies from different angles, backgrounds and conditions.

5.5 Image Features

What is an Image Feature?

An image feature is a characteristic or visual pattern in an image that can help a computer analyse or identify an object.

Features can include structures such as:

• Edges
• Corners
• Shapes
• Patterns
• Textures
• Colours

Types of Image Regions

Region	Description	Usefulness as a Feature
Flat Surface	Large region with little or no variation	Usually difficult to locate precisely
Edge	Boundary between two regions	More useful than a flat surface
Corner	Point where edges meet	Highly useful and distinctive

Flat Surfaces

A flat surface contains very little change in pixel values. Examples include a plain wall or a clear blue sky.

Since many parts of a flat surface look similar, it can be difficult for a computer to identify a specific location on it.

Edges

An edge is a boundary where there is a significant change in intensity or colour.

Examples include the edge of a building, table or window.

Corners

A corner is a point where two or more edges meet. Corners are generally more distinctive than flat surfaces and straight edges.

Example: Library Book Finder

Suppose two books have plain red covers. The red areas contain little variation, so they provide few useful features.

Another book has a title written in a sharp, distinctive font. The letters create many edges and corners, making the book easier to identify.

5.6 Convolution

What is Convolution?

Convolution is a mathematical operation used in image processing and Computer Vision to extract useful information or apply effects to an image.

It involves applying a small matrix called a Kernel or Filter to different parts of an image.

How Convolution Works

An image can be represented as an array of pixel values. A kernel is another small array.

The kernel moves across the image and performs:

1. Element-wise multiplication between the image region and kernel.
2. Addition of the resulting values.
3. Generation of a new output pixel.

What is a Kernel?

A Kernel is a small matrix used during convolution to extract particular features or apply a particular effect to an image.

Different kernels can be designed for:

• Edge detection
• Blurring
• Sharpening
• Other image-processing operations

Padding

When a kernel moves across an image without padding, the output can become smaller because the kernel cannot completely cover the boundary pixels.

Padding adds extra pixels around the boundary of an image so that convolution can be performed near the edges.

A common type of padding uses zeros around the boundary.

Case Study: Blurry Robot Camera

A robot needs to follow a black line on a white floor. Its camera produces a noisy image. The developer applies a blur filter before detecting the line.

The blur filter can be applied using convolution. A suitable kernel averages neighbouring pixel values and reduces unwanted noise.

5.7 Convolutional Neural Network (CNN)

Introduction to CNN

A Convolutional Neural Network (CNN) is a type of Deep Learning neural network particularly useful for analysing images and other visual data.

CNNs can automatically learn important features from images and use these features to classify or recognise objects.

For example, a CNN may learn simple features such as edges in the early layers and progressively learn more complex features such as shapes and objects in deeper layers.

Main Layers of a CNN

1. Convolution Layer
2. ReLU Layer
3. Pooling Layer
4. Fully Connected Layer

1. Convolution Layer

The Convolution Layer applies filters or kernels to the input image to extract useful features.

Initial convolution layers generally detect low-level features such as:

• Edges
• Lines
• Colours
• Simple patterns

Deeper layers can combine these simple features to identify more complex patterns and shapes.

The output produced by convolution is commonly called a Feature Map or Activation Map.

2. Rectified Linear Unit (ReLU)

ReLU stands for Rectified Linear Unit.

It is an activation function commonly used in neural networks. ReLU replaces negative values with zero while retaining positive values.

Input	ReLU Output
-5	0
-2	0
0	0
3	3
8	8

3. Pooling Layer

The Pooling Layer reduces the spatial dimensions of feature maps while retaining important information.

The two common types are:

• Max Pooling – selects the maximum value.
• Average Pooling – calculates the average value.

Pooling Type	Operation
Max Pooling	Selects the largest value from the region.
Average Pooling	Calculates the average of values in the region.

Pooling helps to:

• Reduce the size of feature maps.
• Reduce computational requirements.
• Retain important features.
• Provide some resistance to small changes in position.

4. Fully Connected Layer

The Fully Connected (FC) Layer generally performs the final classification based on the features extracted by earlier layers.

The feature maps can be flattened into a vector and passed to the fully connected layer.

The output can provide probabilities or scores for different classes.

Example: A trained CNN may determine that an image has a high probability of being a "Cat" rather than a "Dog".

CNN Workflow

CNN Layer Summary

Layer	Main Function
Convolution	Extracts features using filters/kernels
ReLU	Introduces non-linearity by removing negative values
Pooling	Reduces spatial dimensions while retaining important information
Fully Connected	Performs final classification

Low-Level vs High-Level Features

Early Layers	Later Layers
Edges	Complex shapes
Lines	Object parts
Basic colours	Complete objects
Simple patterns	Faces, animals, vehicles etc.

Case Study: Animal Shelter App

A student develops an application that identifies dog breeds from photographs. A rule-based system performs poorly when lighting and viewing angles change.

A CNN can learn visual features from training data and can therefore be more suitable for such image-recognition tasks.

The initial convolution layers can detect low-level features such as edges and colours, while deeper layers can learn more complex patterns.

5.8 Python Libraries in Computer Vision

Professional Computer Vision applications often use programming languages such as Python along with specialised libraries.

A library is a collection of pre-written functions and tools that programmers can use instead of creating everything from scratch.

1. TensorFlow

TensorFlow is an open-source framework developed by Google and widely used for machine learning and deep learning.

It can be used for:

• Training neural networks
• Deep Learning
• Image classification
• Computer Vision applications

2. Keras

Keras is a high-level deep learning API that provides a simpler interface for building neural networks.

Keras can be used to create models containing layers such as convolution and pooling layers.

3. OpenCV

OpenCV stands for Open Source Computer Vision Library.

It is an open-source library widely used for Computer Vision and image-processing applications.

OpenCV can work with:

• Images
• Videos
• Webcams
• Real-time visual information

Applications of OpenCV

Application	Possible Use of OpenCV
Facial Recognition	Detecting and analysing faces
Face Filters	Detecting facial landmarks
Self-Driving Cars	Processing camera images and detecting road objects
Retail	Analysing products and customer movement
Medical Imaging	Processing and analysing medical images
Agriculture	Analysing crop images and detecting visual patterns

TensorFlow vs Keras vs OpenCV

Library / Framework	Main Purpose
TensorFlow	Machine Learning and Deep Learning
Keras	Building neural networks using a high-level interface
OpenCV	Image processing and Computer Vision

Case Study: Social Distancing Monitor

A school wants to monitor whether students are maintaining an appropriate distance in a queue. A Python-based system uses a camera to capture video frames and an AI model to identify people.

OpenCV can be used for acquiring and processing the camera frames, while TensorFlow/Keras can be used to build or run a neural network model for visual recognition.

Quick Revision – Unit 5

Concept	Remember
Computer Vision	Enables machines to understand visual information.
OpenCV	Open-source Computer Vision library.
Pixel	Smallest unit of a digital image.
Resolution	Number of pixels represented by image dimensions.
Grayscale	One channel; usually values from 0 to 255.
RGB	Three channels: Red, Green and Blue.
Classification	Identifies the class/category of an image.
Localisation	Identifies the location of an object.
Object Detection	Identifies and locates multiple objects.
Segmentation	Identifies object regions at the pixel level.
Lobe	No-code tool useful for image classification.
Teachable Machine	Google's beginner-friendly web-based AI training tool.
Orange	Visual, open-source data mining and machine learning tool.
Image Feature	Useful visual characteristic such as an edge or corner.
Convolution	Applies a kernel to image regions using mathematical operations.
Kernel	Small matrix/filter used during convolution.
CNN	Deep Learning network designed especially for visual data.
ReLU	Converts negative values to zero.
Pooling	Reduces spatial dimensions while retaining important information.
TensorFlow	Machine Learning and Deep Learning framework.
Keras	High-level API for building neural networks.

Important Differences for Examination

Computer Vision vs Image Processing

Computer Vision	Image Processing
Understands visual information.	Manipulates or enhances images.
Can identify objects.	Can resize, crop or enhance images.
Example: Face detection.	Example: Brightness adjustment.

Classification vs Object Detection

Classification	Object Detection
Identifies the class/category.	Identifies and locates objects.
Usually gives a label for the image.	Can give labels and bounding boxes.
Example: Cat.	Example: Cat + Dog + Bird with boxes.

Grayscale vs RGB

Grayscale	RGB
One channel.	Three channels.
Contains shades of gray.	Contains colour information.
Pixel values generally range from 0 to 255.	Each pixel has R, G and B values.

Convolution vs Pooling

Convolution	Pooling
Uses filters/kernels to extract features.	Reduces spatial dimensions.
Produces feature maps.	Retains important information while reducing size.
Helps detect features such as edges.	Uses operations such as maximum or average.

Competency-Based and Case Study Questions

MCQ 1

Which of the following is an example of Computer Vision?

1. Changing the brightness of an image
2. Cropping a photograph
3. Detecting a pedestrian in a road image
4. Resizing an image

MCQ 2

Which tool is an open-source Computer Vision library?

1. Keras
2. OpenCV
3. Excel
4. PowerPoint

MCQ 3

Which Computer Vision task identifies multiple objects and their locations in an image?

1. Classification
2. Object Detection
3. Image Resizing
4. Image Compression

MCQ 4

What is the maximum pixel value in a standard 8-bit grayscale image?

1. 0
2. 128
3. 255
4. 512

MCQ 5

Which CNN layer reduces the spatial dimensions of feature maps?

1. Pooling Layer
2. Fully Connected Layer
3. ReLU Layer
4. Input Layer

Assertion and Reasoning

Assertion (A): Computer Vision can be used for facial recognition.

Reason (R): Computer Vision can analyse visual characteristics of faces.

1. Both A and R are true and R is the correct explanation of A.
2. Both A and R are true but R is not the correct explanation of A.
3. A is true but R is false.
4. Both A and R are false.

Case Study: Smart Traffic System

A city installs cameras at traffic intersections. The Computer Vision system identifies cars, buses, pedestrians and traffic lights.

Questions:

1. Which Computer Vision task is being used?
2. Why is object detection more suitable than simple classification?

Case Study: Google Translate

A student travels to another country and points a smartphone camera at a sign written in a foreign language. The application recognises the text and displays its translation.

Question: Which technology helps recognise the text?

Case Study: Image Resolution

A photograph has a resolution of 1280 × 720.

Question: How many pixels does the image contain?

Case Study: Smart Waste Sorter

A school uses Teachable Machine to classify waste into paper, cans and bottles. During testing, the system incorrectly identifies some bottles as cans.

Question: What can the school do to improve the model?

One-Minute Exam Revision

Key Terms to Remember

Computer Vision, Image Processing, OpenCV, Pixel, Resolution, Grayscale, RGB, Classification, Localisation, Object Detection, Image Segmentation, Lobe, Teachable Machine, Orange, Image Feature, Edge, Corner, Convolution, Kernel, Padding, CNN, Convolution Layer, ReLU, Pooling, Fully Connected Layer, TensorFlow, Keras, OCR.