COMPUTER VISION

INTRODUCTION

Computer Vision (CV) is concerned with modelling and replicating human vision using computer
software and hardware. In very simple terms, it is an interdisciplinary scientific field that deals
with images. Computer Vision relies on pattern recognition techniques to train computers to
process, analyze, and interpret images. The processing of the image is achieved with the
application of specific algorithms. The basic functions involved in Computer Vision are Text
Extraction (Optical Character Recognition), understanding and classifying Images, Spatial
Analysis etc.

Computer Vision resembles a Jigsaw Puzzle

In a Jigsaw puzzle, different pieces are assembled to form the correct

final picture. The neural network of computer vision also works with
the same approach as solving a jigsaw puzzle. The computer is
provided with a database of thousands of related images and is trained
to recognize specific objects.

Three basic categories of Computer Vision are -

• Low-level vision − Processing images for feature extraction.
• Intermediate-level vision − Object recognition and 3D scene interpretation
• High-level vision − Conceptual description like activity and behaviour of a scene.

APPLICATIONS OF COMPUTER VISION

Autonomous Vehicles

Computer Vision plays a vital role in

autonomous (self-driving) vehicles. It
enables the vehicle to identify objects and
obstacles, and detect traffic signals and
lanes with the help of artificial intelligence
and various sensors like Cameras, Lidar,
Radar etc. Some companies like Tesla and
Yamaha have used the concept of
computer vision in their products.
 Safety & Surveillance

Computer Vision plays a significant role in

a wide range of Safety and Security
measures. The safety and security
applications involve continuous
monitoring using surveillance cameras,
and facial recognition at Airports, Banks,
malls, parking lots, etc. Computer vision
and AI systems provide real-time warnings when employees are detected in hazardous situations
in factories, mines, etc.

Medical / Healthcare
Medical Image Processing is one of the most significant applications of Computer vision. In the
Medical field, Computer vision has immensely facilitated healthcare professionals in extracting
data from medical diagnostic images for accurate and timely diagnosis. It benefits in reducing the
time and efforts required in recognizing health conditions and detecting abnormalities in
Angiography Images, Ultrasonic Images, MRI, CAT scans, X-rays etc., with a higher degree of
accuracy. Computer Vision has been a great help in identifying and spotting various life-
threatening diseases at very early stages.
Microsoft's InnerEye is one of the software that is capable of detecting tumours and other
irregularities in X-Rays. Scan images are uploaded in the software by the Radiologists, and the
software then marks areas for various parts of the organs and ligaments. It then highlights the
areas with tumours which the radiologist can then study to form a proper conclusion.

Manufacturing

The manufacturing industry benefits from Computer Vision in various tasks, including quality
control for products, health and safety, increased production efficiency, and accurate assembly of
components. Manufacturers across every industry use computer vision technology to boost the
productivity and quality of their production lines. Mining companies use computer vision aided
monitoring systems to closely monitor drilling equipment to identify defects and other damage
before they can cause an accident.

WebSPECTOR, a surface inspection system that identifies defects in items, stores images,
and collects image-related metadata to classify errors by type and grade.

Retail
More and more companies functioning in the retail and e-commerce sectors are now using
computer vision solutions to meet customer needs effectively and efficiently. The application of
Computer vision in retail is vast. Cameras installed in retail stores allow retailers to collect large
amounts of visual data assisting in designing a better customer and employee experience, cashier-
less and self-checkout stores, crowd analysis classifying customers' buying habits etc.
Agriculture
Agriculture is considered the economy-boosting
sector of any nation, and with the help of computer
vision, the agriculture sector is achieving a solid base.
Computer vision has improved the agricultural
sector's overall functioning from better productivity
to lower production costs with automation. In
agriculture, computer vision has helped the farmers in
the areas of planting, harvesting, advanced analysis of
weather conditions, weeding and plant health
detection, pesticide spraying and drone-based crop
monitoring.

Marketing

Computer vision gives machines the ability to visualize images through mathematical
representations of three-dimensional shapes and appearances. One can identify subject matter in a
snap and comprehend meaning and context as humans do through computer vision.

Neutrogena delivers personalized experiences to customers with the help of an app. They have a
skin-scanning gadget that can be attached to a smartphone, and the scanner of the smart phone
syncs with the Skin360 app, which analyzes the scans and determines skin health. Neutrogena
products are recommended to the customers after all analysis of all the measures and parameters.

WORKING OF COMPUTER VISION

Computer vision is a field of study of artificial intelligence that aims to develop digital systems,
which help in scanning, interpreting, processing, and analyzing images in the same way humans
do. The model of computer vision is based on training computers to process an image at a pixel
level. Technically, machines attempt to retrieve visual information, handle it, and interpret results
through special software algorithms. With computer vision, machines can accurately recognize
and detect objects, analyze and make meaningful understandings out of a sequence of images.
The operations involved in CV include:
• Procuring the images.
• Processing the data in the images.
• Extracting multidimensional data.
• Analysis of data to find patterns in the data to identify and classify different objects
DIFFERENT TASKS IN COMPUTER VISION
The various applications of Computer Vision are based on extracting information after
processing and interpreting the input image. The extracted information is used for
prediction, and it also helps in forming the foundation for further analysis of the image.
The computer vision applications perform the following tasks:-
• Image Classification :
Image Classification is an essential and primary task that aims at understanding the entire
image and helps in determining the objects present in the image. The objective is to select
the category to which the image belongs.
• Classification + Localization :
The primary task of Classification & Localization is to produce a class label as in image
recognition and also a bounding box that defines the spatial location of the object in the
picture. It is used only for single objects.
• Object Detection :
Object Detection in an image or video data starts with a systematic division of the
picture. At first, algorithms are applied to identify regions of interest in the photo, which
results in the range of object proposals. The final step of detection is to classify objects
 based on models and probability. The results of Object Detection are multiple bounding
boxes and category labels.
• Instance Segmentation :
One of the challenging computer vision tasks is Instance Segmentation, which detects
and defines each distinct object appearing in an image. It detects instances of the things
and assigns a category to the pictures by labelling each image pixel. The segmentation
algorithm generates a pixel-wise mask for each object in the image.

IMAGES
Image is an assembly of PIXELS, and each pixel is like an array of numbers. These numbers
determine the intensity, brightness, and colour of the pixel. An image is a pictorial representation
of an object.

IMAGE FEATURES
Important areas of an image form Image features and are unique for every image. In machine
learning, image features are critical as they form the basis for analyzing, describing and matching
the photos. They are used to train different classifiers and label each object, such as trees, flowers,
children, and birds in the case of a picture of a garden.

PIXELS
Pixels (picture elements) are the smallest items of data in an image. Every digital image is
comprised of pixels. Pixels are arranged in a 2-dimensional grid, i.e. rows and columns. The
number of pixels in an image is also known as its resolution. An image having a resolution of
1024×512 means there are 1024 pixels from left to right (columns) and 512 from top to bottom
(rows). Number of pixels in an image with resolution 1024 x 512 = 5,24,288 pixels. The size of a
pixel (Pixel Density) in an image depends on the pixels per inch (PPI).

PIXEL VALUE
Each of the pixels of an image has a numeric value, known as a pixel value which defines the
brightness, intensity and colour of the pixel etc. The most common pixel format is the byte image,
where this number is stored as an 8-bit number giving a range of possible values from 0 to 255.
Zero is the value used for no colour or black, and full colour or white is represented by 255.

Why do we have a value of 0 to 255 ?

In computer systems, every bit (binary digit) stores a value either 1 or 0. Each pixel uses 1 byte of an
image, which is equivalent to 8 bits of data. Each binary digit can have only two possible values (0 or 1 ),
so each 8-bit number forms 28 = 256 unique combinations, which start from 0 and end at 255.

2.5 RESOLUTION

Image resolution is the measurement of an image in the form of horizontal x vertical pixels. So, an
image resolution of 600 x 400 tells us that the image measures 600 pixels in width and 400 pixels
in height. Image resolution represents the detail an image holds; higher resolution of the image
means higher detailing of the picture. Image resolution is expressed in PPI (Pixels per inch)

2.6 GRAYSCALE IMAGES

• The size of each pixel in a Grayscale image is 1 byte (8 bits)

• Each pixel can take values from a range of 0 to 255.
• Each value represents shade between black and white (0 for black and 255 for white and
in between different shades of greys)
• The number of channels used in a grayscale image is 1
• The depth of a grayscale image is 8 bits
• The size of a grayscale image is defined as the Height x Width of that image.
 Example of Grayscale Image

2.7 RGB Images

• Each pixel in RGB images stores three values for three primary colours – Red, green and
Blue:
Red : 0-255
Green : 0-255
Blue : 0-255
• Different concentrations/intensities of Red, Green and Blue colours combine to form other
colours in the image.
• The depth of an RGB image is 1 byte, i.e. 8 bits
• The number of channels for an RGB image is 3

For example, RGB (Colour) image in JPG form and its resolution is 512 x 512. The dimensions
of the array used to represent the image will be 480 x 480 x 3. Each element of the array describes
the pixel intensity and has a value ranging between 0 to 255.

RGB Model Processing of RGB

 Each kind of image has a few attributes attached to it, like several channels and depth.
· Number of Channels: Defines the dimension of the array
· Depth: The maximum bit size of the number which is stored in the array

3.1 INTRODUCTION TO OPENCV

Open Source Computer Vision, abbreviated as OpenCV, is a python
library extensively used for image recognition or identification. The
library was officially launched in 1999 by Intel, and in the early stage, it
was written in C/C++, but now it is commonly used in Python for computer
vision.
OpenCV is a massive open-source, cross-platform library used for
Computer Vision in Artificial Intelligence, Image Processing, Facial
recognition, Object Detection etc.

Features of OpenCV library :

• Image processing and analysis
• Processing of images like filtering, transforming and rotating etc.
• Detect specific objects such as faces, eyes, and cars in the videos or pictures.
• Analyze the video, i.e., estimate its motion, subtract the background, and track objects in
it.
To install OpenCV library, open Anaconda prompt and write the command :
pip install opencv-python