Image Processing with Python

Sampriti Chatterjee (Great Learning)

Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
 Agenda

1 What is Image Processing?

7 Convolutional Layer

2 Traditional ML VS Deep learning

8 Relu layer

3 Human brain VS AI
9 Pooling layer

4 How Image Input work in computer?

10 Fully connected layer

5 What is CNN?
11 Demo: Image Processing with Python
6 Layers in CNN

Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
 What is Image Processing

Image processing is a method to perform some important operations on

an image. In order to get an enhanced high Quality image or to extract the
most useful information from that

 It is a one type of signal processing

 In this processing input is an image and output may

be image or characteristics/features associated with

that image

Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Is It A Cat Or A Dog?

Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
 Is It A Cat Or A Dog?
Traditional Machine Learning Deep Learning

Manual selection from so many This is where traditional machine

features is impossible learning fails and deep learning comes
into picture
Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Human Brain VS Neural Network

Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
 Human Brain VS Neural Network

Neuron Artificial Neuron

• Cell Nucleus • Node

• Dendrites • Input: x1, x2

• Synapse • Weights or interconnections

• Axon • f(x) (mapping, activation, learning)

• Terminal Axon • Output: y1

Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
 How Image Input Works In Computer?

4 21 54 92 48
Converts the image into an array of pixel
values where the dimension of array 2 22 54 36 22
depends on the resolution of the image 3 42 37 86 73

Array of dimension 32 X 32 X 3 (The 3

refers to RGB values)

Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
 What is CNN?
“A feed forward network to process and recognize image data with the
grid version”

Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Fully Connected Network
Combines the extracted features
and represents a new model

Pooling Layer
Reduce the image size

ReLu Layer
Converts negative numbers into zero

Convolution Layer
Converts images into an array
Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
 Layers In CNN

Layers in CNN
Convolutional Layer

“Converts images into an array”

1 Convolutional Layer

2 ReLU  First layer of CNN

 Stores the pixelated values of image into an array

3 Pooling Layer
 Used for extracting the features of the image and reducing
4 Fully Connected Layer its dimensionality

Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Working Of Convolutional Layer
4 21 54 92 48 4 21 54 92 48
0 -1
2 22 54 36 22 2 22 54 36 22 * 1 1
3 42 37 86 73 3 42 37 86 73 Filter/Feature detector

Terminology Alert!!

3 22 -2 10
23 25 87 139
 Extracted features from the image
 Dimension reduced

Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
 Layers In CNN
Activation function: ReLu
Layers in CNN
“Converts negative values into zero”

1 Convolutional Layer  ReLU is a half rectifier

 f(y) = 0 when y < 0
2 ReLU
 f(y) = y when y > = 0
 Range of ReLU : [0 to infinity]
3 Pooling Layer

4 Fully Connected Layer

ReLu

Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Working Of Relu Layer
4 21 54 92 48
3 22 -2 10
2 22 54 36 22 Applying ReLu layer
23 25 87 139
3 42 37 86 73

3 22 0 10
After removing the negative values
23 25 87 139

Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
 Layers In CNN
Pooling Layer
Layers in CNN
“Reduces the spatial size and the numbers of parameters”

1 Convolutional Layer
 Used to reduce dimensionality
 Helps to control overfitting
2 ReLU
 Filters of size 2x2 are commonly used in it

3 Pooling Layer

4 Fully Connected Layer

Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Working Of Pooling Layer
3 22 0 10 Max Pooling 3 22 0 10
2*2 with stride = 1
23 25 87 139 23 25 87 139

Terminology Alert!!
Amount of movement between applications of the filter to the 25
input image is referred as stride

3 22 0 10 3 22 0 10 3 22 0 10

23 25 87 139 23 25 87 139 23 25 87 139

25 87 25 87 139 25 87 139 139

Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Flatten The Data
“Converting the Pooled feature map into an array is known as data flattening”

25
87
25 87 139 139 Flatten
139

139

Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
 Layers In CNN
Pooling Layer
Layers in CNN
“Combines all the features together to create a final model”

1 Convolutional Layer
 Used to reduce dimensionality
 Helps to control overfitting
2 ReLU
 Filters of size 2x2 are commonly used in it

3 Pooling Layer

4 Fully Connected Layer

Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Working Of Fully Connected Layer
Hidden layers
Input layer

25 Prediction

87 Dog

139 Cat

Output layer
139

Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Image Processing with Python
Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
 Thank You

Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited
Proprietary content. ©Great Learning. All Rights Reserved. Unauthorized use or distribution prohibited