MACHINE LEARNING LAB 3

Introduction to OpenCV, PIL

MUNADI SIAL

SCHOOL OF ELECTRICAL ENGINEERING AND COMPUTER SCIENCE

NATIONAL UNIVERSITY OF SCIENCES AND TECHNOLOGY
Overview
This lab will be centered on the following:
• Understand the image data as an array of pixels
• Load and Save images
• Display images in different windows
• Access and modify pixels in images
• Access and modify regions in images
• Place lines, rectangles, circles and text in images
• Resize images at various scales
• Rotate images at various angles
OpenCV
• OpenCV is an imaging library used widely for computer vision tasks such as
feature extraction, stereovision, image stitching, pose estimation and many
more

• Computer vision is the field that involves getting information from image data

• OpenCV is available in Python, C++ and MatLab

• To use OpenCV in Python, we import the library:

import cv2
Images
An image is an array of unit squares called “pixels”

Pixels

Each pixel has a location (x,y)

in the image array

Each pixel has some color

values (intensity level)
Images
A colored image has 3 channels, i.e. there are 3 color values in every pixel:
• Blue (channel 0)
• Green (channel 1)
• Red (channel 2)

By default, OpenCV stores channels as BGR (not RGB)

Images
An image is essentially a 3-D array defined by its height (y), width (x) and number
of channels (n)

Notice that the y-axis points n

downwards in the pixel frame

y
Load Images
To load an image, we use the imread function:

imgA = cv2.imread(“my_image.jpg”, 1)

The above function will load the image file and place it in imgA.

The imgA is like a variable that stores an image and we can use it in our program to
reference the image

(The second argument in the imread function chooses the color mode;
0 is for grayscale, 1 is for colored and -1 is for alpha-channel inclusive)
Load Images
If the image is in the same directory as the script file, the image filename is directly
used:

imgA = cv2.imread(“p3at.jpg”, 1)

If the image is in a different directory then the entire path to the image must be
provided:

imgB = cv2.imread('D://docs/Robot Pics/turtlebot.jpg', 1)

Display Images
Once the image is loaded, it can be displayed using the imshow function:

cv2.imshow('image1',imgA)
The first argument is the window name. This name appears at the top of the window
which shows the image.

The second argument is the image object that is to be displayed

Display Images
Once the image is loaded, it can be displayed using the imshow function:

cv2.imshow('image1',imgA)
cv2.waitKey(0)
When the imshow function is used, the image appears only for a very small time
before it closes. To avoid this, the waitKey function is used to hold the image.

The argument for the waitkey function specifies the time (in milliseconds) to hold
the image. If the argument is 0, then the image is held for infinite time until the user
presses a key.
Display Images
The following code will display two images
cv2.imshow('image1',imgA)
cv2.imshow('image1',imgB)
cv2.waitKey(0)
cv2.destroyAllWindows()

To ensure all windows are closed, the destroyAllWIndows function is used.

(Another function called destroyWindow(‘window_name’) can be used to close a

specific window)
Save Images
To save an image on disk, the imwrite function is used:

cv2.imwrite(‘myImageRotated.jpg',imgA)

The first argument is the name of the image file to be saved

The second argument is the image object that we are saving

Basic Operations
Consider the image of size 620 x 420

The x-axis goes from 0 to 619

The y-axis goes from 0 to 419
0 619
(The channels go from 0 to 2)
0
x

419
y
Basic Operations
To get the BGR values of a specific pixel at location (px, py), we use:

img = cv2.imread(‘bird.jpg', 1)

val = img[py, px, :]

0 px 619

The first index (py) is the row of the

0
x
array. It corresponds to the y-axis

The second index (px) is the column of

the array. It corresponds to the x-axis
py (px,py)
The third index (:) is the channel
419
y
Basic Operations
To get the BGR values at a number of pixels:

img = cv2.imread(‘bird.jpg', 1)

print(img[380,62,:])
0 619
print(img[150,150,:])
print(img[48,350,:])
print(img[372,588,:])
0
x
(350,48)

(150,150)
Output: [27 59 42]
[ 2 54 251]
[251 151 127] (62,380) (588,372)
[ 0 102 184] 419
y
Basic Operations
To get the BGR values at location (0,0) pixel:

img = cv2.imread(‘bird.jpg', 1)

print(img[0,0,:]) # BGR
0 619
print(img[0,0,0]) # B
print(img[0,0,1])
print(img[0,0,2])
#
#
G
R
0
(0,0) x

Output: [22 143 63]

22
143
419
63
y
Basic Operations
The shape attribute is used to get the dimensions of the array:

img = cv2.imread(‘bird.jpg', 1)
rows = img.shape[0]
cols = img.shape[1]
0 619
print(img.shape)
print(rows)
0
x
print(cols)

Output: (420, 620, 3)

420
620 419
y
Basic Operations
The following code will load an image and display its color channels

img = cv2.imread(‘bird.jpg', 1)

imgB = img[:,:,0]
imgG = img[:,:,1]
imgR = img[:,:,2]

cv2.imshow('Original',img)
cv2.imshow('Blue',imgB)
cv2.imshow('Green',imgG)
cv2.imshow('Red',imgR)
cv2.waitKey(0)
cv2.destroyAllWindows()
Basic Operations
The results are shown.

The more white color

indicates large values of the
channel color

Note that the color

channels are ordered as
BGR (instead of RGB) in
OpenCV:

imgB = img[:,:,0]
imgG = img[:,:,1]
imgR = img[:,:,2]
Cropping Images
Cropping a region of image (ROI) is similar to accessing pixels
A range of pixel values can be taken using a slice operation (:)
img = cv2.imread(‘bird.jpg', 1)
img_crop = img[50:250,100:400]
cv2.imshow(‘ROI', img_crop) 100 400 619
cv2.waitKey(0)
50 x

250

419
y
Scaling Images
The resize function is used to change the image size
imgResize = cv2.resize(img,None,fx=0.5, fy=0.5,
interpolation = cv2.INTER_CUBIC)

cv2.imshow('Img', img)
cv2.imshow('Img1', img_reSize)

fx and fy are the scaling

factors

Scaling factor of 0.5 means

half of original length
Scaling Images
If the scaling factors fx and fy are not equal, the image gets stretched
img_resized1 = cv2.resize(img,None,fx=1,
fy=0.4, interpolation = cv2.INTER_CUBIC)

img_resized2 = cv2.resize(img,None,fx=0.3,
fy=1, interpolation = cv2.INTER_CUBIC)
Rotating Images
Rotated at 30 degrees
To rotate the image, 2 functions are used:
• getRotationMatrix2D to compute the
transformation matrix
• warpAffine to apply the rotation
transformation to the image matrix

img = cv2.imread(‘bird.jpg', 1)
rows = img.shape[0]
cols = img.shape[1]

M = cv2.getRotationMatrix2D((cols/2,rows/2),30, 1)
rot = cv2.warpAffine(img,M,(cols,rows))

cv2.imshow(‘Rotate', rot)
Rotating Images
The getRotationMatrix2D takes the following inputs:
• Center of rotation (cx, cy)
• Angle of rotation A
• Scale factor k

M = cv2.getRotationMatrix2D((cx, cy), A, k)
Placing Shapes
Various shapes can be placed in the image:
img1 = cv2.line(img,(x1,y1),(x2,y2),(B,G,R),t)
img2 = cv2.rectangle(img,(x1,y1),(x2,y2),(B,G,R),t)
img3 = cv2.circle(img,(xc,yc), radius, (B,G,R), t)
(x1,y1) start point
(x2,y2) end point
(xc,yc) center point
(B,G,R) color values from 0 to 255
thickness line thickness (-1 for solid color)

Text can also be placed in the image:

cv2.putText(img,'Hello', (x,y), cv2.FONT_HERSHEY_SIMPLEX,
size, (B,G,R), thickness, cv2.LINE_AA)
Placing Shapes
The following example places some shapes and text in the image
img = cv2.line(img,(50,50),(600,400),(255,255,0),5)
img = cv2.rectangle(img,(400,50),(550,200),(0,255,255),5)
img = cv2.circle(img,(80,200), 30, (255,0,255), 5)

font = cv2.FONT_HERSHEY_SIMPLEX

cv2.putText(img,'Hello',
(50,350), font, 3,
(255,255,255), 5, cv2.LINE_AA)

cv2.imshow(‘Shapes', img)
Lab Tasks

 Download the manual from LMS

 Perform the Lab Tasks as given in the manual and submit it on LMS
 Remember to execute scripts with the terminal