Computer Vision

Lec 02
What is image?
Image is collection of pixels.
What is pixel?
Pixel is smallest unit in an image which contain color value.
● A good image contain approximately 300 PPI(Pixel Per Inch)
OR DPI(Dot Per Inch)
What is video?
Video is collection of images.
What is image?

An image may be defined as a two dimensional function f(x,y)

where x and y are spatial coordinates and amplitude of f at
any pair of coordinates (x,y) is called the intensity or
Graylevel of the image at that point.
Types of Image

Image

Single channel 3-channel image

Grey Scale RGB(Red Green

Black white image Blue)
image
Representation of Digital Image
● Image is represented in 2-D function f(x,y)
● x,y are spatial coordinate
● f(x,y) show the intensity or grey level of image at that point
● f(x,y) are all finite and discrete quantities are known as digital image

Suppose size of image is(4,5) {width,height} width

f(0,0) f(1,0) f(2,0) f(3,0)

f(0,1) f(1,1) f(2,1) f(3,1)

height
f(0,2) f(1,2) f(2,2) f(3,2)

f(0,3) f(1,3) f(2,3) f(3,3)

pixel
f(0,4) f(1,4) f(2,4) f(3,4)
Image Formats
Suppose size of image is(100,150) {width,height}

Format No. of pixels No. of bits Grey Level

Black white 100*150 100*150*1 (0 or 1)= 2

Grey scale 100*150 100*150*8 (0-255) = 2^8

Color image 100*150*3 100*150*3*8 (0-255)=2^24

import cv2
#cv2.imread(path,flag)
img1=cv2.imread("1.jpeg")
img1=cv2.resize(img1,(3,4)) #3=width(column(x), 4=height(row(y))
print(img1)
print("pixel intensity of x=1,y=1",img1[1,1])
print("shape",img1.shape) x(column)=0,y(row)=0
cv2.imshow("picture",img1)
cv2.waitKey()
[
[ [128 68 159] [122 91 218] [ 36 104 73] ] x(column)=0,y(row)=0
[ [ 65 101 107] [173 234 231] [ 70 137 92] ]
[ [ 9 8 203] [ 46 74 72] [ 90 67 201] ]
[ [ 48 120 73] [ 9 34 96] [ 46 82 89] ]
]
pixel intensity of x=1,y=1 [173 234 231]
shape (4, 3, 3) #4=rows,3=col,3=channel
import cv2

#cv2.imread(path,flag)

img1=cv2.imread("1.jpeg",0) # 0 represent grayscale image

img1=cv2.resize(img1,(3,4))#3=width(column(x), 4=height(row(y))

print(img1)

print("pixel intensity of x=1,y=1",img1[1,1])

cv2.imshow("picture",img1)

cv2.waitKey() x(column)=0,y(row)=1
[
[103 132 87]
[ 99 226 116]
[ 66 70 110]
[ 98 49 80]
]
pixel intensity of x=1,y=1 226
Sampling and Quantization
An image may be continuous with respect to the x- and y-coordinates, and also in
amplitude.
To digitize it, we have to sample the function in both coordinates and also in
amplitude.
● Digitizing the coordinate values is called sampling.
○ It determines the spatial resolution of the digitized images.
● Digitizing the amplitude values is called quantization.
○ It determines the number of grey levels in the digitized images.
Quantization Level

● 8 bit quantization: 28=256 gray levels (0: black, 255: white)

● Binary (1 bit quantization):2 gray levels (0: black, 1: white)
Gray Scale Values (Range 0-255)
Pixels Neighborhood

Any pixel can be represented by

x=column y
y=row
y+1 x-1,y+1 x,y+1 x+1,y+1

x-1,y x,y x+1,y

y
x-1,y-1 x,y-1 x+1,y-1
y-1
x
x-1 x x+1
N₄(p) Neighborhood
Any pixel can be represented by

x=column y
y=row
y+1 x,y+1

x-1,y x,y x+1,y

y
x,y-1
y-1
x
x-1 x x+1
NٖD(p) Neighborhood
Any pixel can be represented by

x=column y
y=row
y+1 x-1,y+1 x+1,y+1

x,y
y
x-1,y-1 x+1,y-1
y-1
x
x-1 x x+1
Nٖ8(p) Neighborhood
Any pixel can be represented by

x=column y
y=row
y+1 x-1,y+1 x,y+1 x+1,y+1
Union of
x-1,y x,y x+1,y
y
Nٖ4(p) and NٖD(p)
x-1,y-1 x,y-1 x+1,y-1
y-1
x
x-1 x x+1
 ADJACENCY
● Let V be the set of intensity values used to define adjacency
4-adjacency

Two pixels p and q with values from V are 4-adjacent if

● q is in the set Np(4).

In a binary image, V = {1} if we are referring to adjacency of pixels with value 1

8-adjacency
Two pixels p and q with values from V are 8-adjacent if q is in the set N8(p).

54 10 58 8

81 150 2 34

201 200 3 45

7 70 147 56

V={1,2,3,....10}
m-adjacency
Two pixels p and q with values from V are m-adjacent if
● q is in N4(p)
OR
● q is in ND(p) and the set N4(p) ∩ ND(p) has no pixels whose
values are from V.
Fig : A

Fig : A Fig : B Fig : C

Fig : D
 Distance Measure
For pixels p, q with coordinates (x,y) , (u,v)

● Euclidean Distance: De(p, q) = [(x-u)2 + (y-v)2]1/2

● City Block Distance: D4(p, q) = |x-u| + |y-v|
● Chess Board Distance: D8(p, q) = max(|x-u|, |y-v|)
City Block Distance

● The pixels with D4 distance ≤ 2 from (x, y) (the center point) form the following
contours of constant distance:
● D4(p, q) = |x-u| + |y-v|
● The pixels with D4 = 1 are the 4-neighbors of (x,y)
1

3
Chess Board Distance

● The pixels with D8 distance ≤ 2 form the following contours of constant

distance
● D8(p, q) = max(|x-u|, |y-v|)
● The pixels with D8 = 1 are the 8-neighbors of the pixel at (x,y)
ARITHMETIC OPERATIONS

Arithmetic operations between two images f(x,y) and g(x,y) are denoted as

Addition : s(x,y) = f(x,y) + g(x,y)

Subtraction: s(x,y) = f(x,y) - g(x,y)

Multiplication: p(x,y) = f(x,y) * g(x,y) (element wise multiplication)

Division: t(x,y) = f(x,y) / g(x,y)

Addition
0 100 10 10 250 5 10 255 15

4 0 10 + 2 0 0 = 6 0 10

8 0 5 0 10 10 8 10 15

Usage:

● Averaging images for noise removal

Image Averaging for Noise Reduction
Subtraction
0 +
100 =
10 10 100 5 0 0 5

4 0 10 2 0 0 2 0 10

8 0 5 0 10 10 8 0 0

Usage:
● Removal of background from images
● Image enhancement
● Image matching
● Moving/displaced object tracking
Removal of background from images
 Multiplication and Division

● Multiplication of pixel values by a number greater than one will

brighten the image
● Division by a factor greater than one will darken the image

a) original image b) image multiplied by 2 c) image divided by 2

Example of Image Multiplication
Multiplication of image can be used for superimposing texture on an image
Logical Operator
AND operator
● Logical AND operations are useful for the masking of images.
● binary image acts as a mask removes information from certain parts of the
image.
● This operation can be used to find the similarity white regions of two different
images (it required two images).
AND (^)
OR Operator
Logical OR operations are useful for the

● compositing of images
 Not Operator
negative of the original image.
Neighbourhood Operation
 Image Enhancement
● is a process to improve the quality of image.
● To highlight the important detail
● To remove noise
Method of image enhancement
● Spatial domain
● Frequency domain
● Combination method
Spatial Domain

● Spatial domain refers to the image plane itself and are

based on direct manipulation of pixels in an image.
● g(x,y) = T(f(x,y))
Method:
○ Point Operations (Intensity Transformation)
○ Histogram based processing
○ Mask Processing
Point Operation
● Enhancement at any point in an image depend on gray level at point.
● S=T(r)
Where
S=value of pixel after processing
T= transformation function
r = value(intensity) of pixel before processing
Method:
○ Contrast Stretching Function
○ Linear Negative Transformation(Image Negative)
Image Negative
Obtain the digit negative of the following 8-bit per
pixel image

139 205 105

141 252 99

201 15 76
 Obtain the digit negative of the following 8-bit per
pixel image
116 50 150
L=28 = 256 139 205 105

S=(L-1)-r 114 3 156

141 252 99

r=p(x,y) 201 15 76 54 240 175

Home Task
● Log transformation
● Power law transformation
Mask/Kernel Processing

Instead of taking 1 pixel, we are taking the neighbourhood values.

N4P - 4 Neighbours : (x+1,y),(x-1,y),(x,y+1),(x,y-1)

NDP - 4 Diagonal Neighbours : (x+1,y+1),(x+1,y-1),(x-1,y+1),(x-1,y-1)
N8P - 8 Neighbours : N4P + NDP

Min, Max, Median operation

 MASK PROCESSING, MASK FILTERING, SPATIAL
FILTERING
● the terms template, kernel, window, filter, and convolution mask are also used
to denote a mask
● mask is an array of numbers (typically 3x3 or 5x5, etc.)
● mask is usually very small with respect to the processed image
● mask values are called coefficients
● values of coefficients determine the nature of the process

the masks can be designed for noise reduction, edge detection, etc.
 Smoothing Filters
● Smoothing filters are used for blurring (removal of small details from
an image prior to large object extraction) and for noise reduction
Low pass Filter:
● Low pass filter always has positive coefficients
● Sum of the coefficients is equal to 1
● It blurs edges
Low Pass filtering
Simple average filters:
Median filtering
● if the objective is to achieve noise reduction rather than blurring, use median
filter.
● less
 Spatial image Sharpening
Goals: highlight or enhance details in images.
● To highlight fine detail
● Remove blurring from image
● Highlight edges
Procedure: Based on spatial differentiation.
Applications:
● Photo enhancement
● Medical image visualization
● Industrial defect detection
Spatial Differentiation
● Rate of change of function
● Consider 1-D example
1st derivative
● Must be zero in flat segments (area of constant grey levels)
● Must be non zero at the onset of grey level step or ramp
● Must be non zero along ramps
2nd Derivative
● Must be zero in flat segment
● Must be non zero at the onset and end of a gray level step or ramp
● Must be non zero along ramps of constant slope
1st vs 2nd Derivative comparison

● Must be zero in flat segments ● Must be zero in flat segment

(area of constant grey levels)

● Must be non zero at the onset ● Must be non zero at the onset
of grey level step or ramp and end of a gray level step or
ramp
● Must be non zero along ramps ● Must be non zero along ramps
of constant slope
f(x,y) g(x,y)
 Unsharp Mask & High Boost Filtering
Using sequence of linear Spatial filters in order to get sharpening effect.

● Blur
● Subtract from original image
● Add resulting mask to original image
A=K
Horizontal Filter:
Vertical Filter
1st Derivative Filter

● Robert
● Sobel
● Prewitt Filter
Roberts cross-gradient operators.

3*3 region of image

Prewitt Mask