Denoising Autoencoder

Table of conents

• Description
• Used libraries
• Model architecture
• Graphs

Description

This project has been made to create my very first autoencoder and to practice.
The used dataset is a classic MNIST digit dataset, available in built-in datasets in keras.
The dataset contains 28x28 images of digits. I added random noise to each image and tried to reconstruct it by using an autoencoder.
The encoded images are 7x7 images.

Used libraries

tensorflow==2.5.0
numpy==1.19.5
matplotlib==3.4.1

Model architecture

• Encoder
  
• Decoder
  

Graphs

• Clean vs noisy data
  
• Loss plot
  
• Final results
  
  The noise factor was equal to 0.4
  The first row is the clean test data
  The second row is the noisy test data
  The third row is the encoded noisy test data
  The last row is the decoded noisy test data
• Results for noise factor = 0.6
  
• Results for noise factor = 0.8
  
• Results for noise factor = 1.0