Leveraging CNNs in Auto Encoder Architecture to remove noise from scanned documents and get encoder - decoder pair models for encoding and decoding clean documents.
This project aims to showcase the practical use of Autoencodersfor denoising documents while leveraging their inherent capacity for image encoding. My model, constructed using Convolutional Neural Networks within the Autoencoder Architecture, is trained on a dataset provided by RM.J. Castro-Bleda, S. España-Boquera, J. Pastor-Pellicer, F. Zamora-Martinez, available through the UCI machine learning repository.
We will also explore the Vanilla Auto Encoder for the same purpose.
The model's objective is to remove or reduce noise found in textual documents, such as watermarks and wrinkles, and provide clean versions along with their corresponding encoded images.
