How can we use neural networks to denoise images?
Furthermore, how good are the U-Net and Autoencoder models at denoising images?
Image noise is random variations of brightness or colors in images. Noise can be caused by a number of factors, like: poor lighting conditions, a high ISO (a camera's sensitivity to light), long exposure times, heat, and more.
In this study, Alex and I tested the U-Net and Autoencoder models for Image Denoising—a task in computer vision problem that seeks to remove noise from an image in order to yield a cleaner result.
We implemented our own U-Net and Autoencoder models (our code can be found in image-denoising.ipynb), and compared them qualitatively and quantitatively: we used the MNIST dataset to evaluate the denoising performance for both networks. Our results can be found in our report: Image_Denoising.pdf
The architecture implementations are laid out in the report, as well as the rationale behind them
Testing the models' ability to remove noise using different noise levels inputs
The report lays insights on what the most effective/efficient strategies are for this task
Quantitatively measuring both models' performances on denoising images
