Image Denoising

Overview

Deep learning(DL) is an area of machine learning that deals with artificial neural networks, which are algorithms inspired by the structure and function of the brain.
Autoencoder(AE): An Autoencoder is a type of artificial neural network that has bottleneck architecture that turns a high-dimensional input into a latent low-dimensional code (encoder), and then performs a reconstruction of the input with this latent code (the decoder).
Denoising Autoencoder(DAE): is a modification of the autoencoder. Denoising autoencoders corrupt the input data, adding noise to the input image, and then try to reconstruct the original image from the noisy image.

With the increase in digital images taken every day the demand for more accurate and pleasing images has also increased.
Image Denoising will help the above task as the name itself suggests we will remove noise from the noisy image to restore the original image.
We at IvLabs gave a try at this task of Image Denoising with the help of Convolutional Denoising Autoencoder.

Approach:

The denoising autoencoder is implemented with PyTorch and is applied on the MNIST and Fashion MNIST datasets.
The encoder network consists of three convolutional layers while the decoder network has three trans convolutional layers.
Encoder downsamples the data and then decoder reconstructs the original data from the lower-dimensional representation.
Gaussian noise with a noise factor value of 0.5 is added to distort the images in the datasets.
Each layer has a ReLU activation function and the final layer of the decoder has a sigmoid activation function.
MSE loss and adam optimization was used for updating the parameters.