Residual Learning of Deep CNN for Image Denoising

Image denoising is a process of taking a noisy observed image with the aim of restoring the clean image while minimizing the damage of the original image properties.
This process is an important component of many applications.
In recent years, deep convolution neuronal networks (CNN) have been widely used to achieve this goal, one being the DnCNN network which focuses on Additive White Gaussian Noise (AWGN) while the real-world noisy patterns is a signal that cannot be described by an explicit distribution model.
In this project we are exploring several ways to improve the proposed network by changing the loss function used for network training, training the network on a known image dataset (SIDD) containing images with real noise, when we assume nothing about the noise distribution in the image. We will examine the improvements compared to several leading works in the field.

Residual Learning of Deep CNN for Image Denoising