Scribd
Upload
2 views3 pages

CNN Image Processing Explained

This document explains the step-by-step process of how a Convolutional Neural Network (CNN) processes a 6x6 grayscale image. It covers the input image, application of a convolution filter, ReLU activation, max pooling, flattening the output, and using fully connected layers for classification. Visualizations are included to enhance understanding of each stage in the CNN workflow.

Uploaded by

hewan bekele
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
2 views3 pages

CNN Image Processing Explained

This document explains the step-by-step process of how a Convolutional Neural Network (CNN) processes a 6x6 grayscale image. It covers the input image, application of a convolution filter, ReLU activation, max pooling, flattening the output, and using fully connected layers for classification. Visualizations are included to enhance understanding of each stage in the CNN workflow.

Uploaded by

hewan bekele
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

Understanding CNNs with Visuals

Step-by-Step Example Using a 6x6 Image

Introduction
In this document, we walk through how a Convolutional Neural Network (CNN) processes an image
step by step, using a realistic 6x6 grayscale image and showing all calculations clearly.
We also include visualizations using the TikZ package to help you see what’s happening at each stage.
—

Step 1: Input Image (6x6 Grayscale)


Let’s begin with a small grayscale image of size 6 × 6:
 
0 0 0 0 0 0
0 5 5 5 5 0
 
0 5 9 9 5 0
Input Image = 0

 5 9 9 5 0
0 5 5 5 5 0
0 0 0 0 0 0
Below is a visual representation using shades of gray:

This could represent part of a digit or shape.


—

Step 2: Convolution Layer – Apply Filter


Apply a vertical edge detection filter:

1
 
−1 0 1
Filter (3x3) = −1 0 1
−1 0 1
Now slide this filter over the image and compute dot products.
Here’s one position visualized:

Dot product calculation:


(−1)(0)+(0)(0)+(1)(0)+(−1)(0)+(0)(5)+(1)(5)+(−1)(0)+(0)(5)+(1)(9) = 0+0+0+0+0+5+0+0+9 = 14
Continue this process across the image...
Final Feature Map:  
14 −4 −14 0
10
 0 −20 0
10 0 −20 0
14 −4 −14 0
—

Step 3: ReLU Activation – Remove Negatives


Apply ReLU(x) = max(0, x):
Before ReLU:    
14 −4 −14 0 14 0 0 0
10 0 −20 0 10 0 0 0
  ⇒ After ReLU =  
10 0 −20 0 10 0 0 0
14 −4 −14 0 14 0 0 0
—

Step 4: Max Pooling – Reduce Size


Use a 2 × 2 window and stride = 2.
Result after max pooling:  
14 0
14 0
Visualized:

2
Step 5: Flatten – Turn into a List
Convert 2D output to 1D:

Flattened Output = [14, 0, 14, 0]


—

Step 6: Fully Connected Layers – Make Prediction


The flattened vector ‘[14, 0, 14, 0]‘ is passed to dense layers which combine features to classify the image.
Output might look like:
[0.8, 0.1, 0.05, 0.05] ⇒ Predicted Class: 0

You might also like