Old Degraded Document Restoration using Deep Learning via Generative Networks
We are living in the era of digitization. The data collection and documentation processes are taking place digitally now. However, the same cannot be said about the past. In the past, documentation was carried out using hand-written text and maintained on hard copies. This made it vulnerable to diff
2025-06-28 16:28:41 - Adil Khan
Old Degraded Document Restoration using Deep Learning via Generative Networks
Project Area of Specialization Artificial IntelligenceProject SummaryWe are living in the era of digitization. The data collection and documentation processes are taking place digitally now. However, the same cannot be said about the past. In the past, documentation was carried out using hand-written text and maintained on hard copies. This made it vulnerable to different types of degradations overtime, which make them hard to read and understand even via modernized applications such as OCRs. Our project is concerned with such old documents which have been corrupted by degradations and aims to remove these unnecessary features to retrieve a clean version of the document without changing its original content. The main component of our project will be a pre-trained Generative Network that will be available for use by anyone using the Web application. The user would upload the document via a web interface and after some processing, receive a clean version of the document as output.
Project Objectives- An easily accessible web-platform for different kinds of users and organizations.
- A software that allows document-to-document conversion by removing degradations from the provided document (if any) and returning a cleaner version of the document without modifying the original document-text.
- An application that also allows the user to preserve the documents in a secure-account in case of any sort of misplacement.
The overall web application system would comprise of three major components i.e. the Front-end GUI, the Backend web interface and the Generative Adversarial Network (GAN) Model to remove degradations from document. The Front-end GUI will be implemented using Angular Framework. On backend, we will use Flask for easier integration with Frontend. The Python Imaging Library (PIL) will be also used to convert the document received from frontend into a set of images. The Generative Network model will receive these images to remove any degradations that might exist in the document. Finally, the clearer version of the images will be re-compiled and a document would be returned in PDF format.
Benefits of the ProjectThe outcome of the project is an online web-tool that takes a degraded document as input and provides a cleaner version of the document by removing basic degradations. Some possible use-cases for our application may include an example in which a person wants to restore and preservesome old documents which may have been distorted by wear and tear. Another example could be a digital library which is trying to maintain an archive of historic manuscripts and letters. They could use this application to get clean version of the document.
Technical Details of Final DeliverableThe technical details of the final delivrable are as follows:
Extraction:
This part of the process is based on a simple PDF rendering library that allows to take in a PDF document and convert it into a series of images, each representing a single page from the document. Hence, the output of the component is a set of degraded document images to be restored to their original condition.
Adversarial Model:
The Adversarial Model is a pre-trained model to be used for removal of degradations from the document image. The model will be trained on a common dataset for degraded document images and will make use of binarization method to remove noise and preserve the original text of the document. Furthermore, since Generative Adversarial Networks (GANs) have proven to the best technique for image generation task over the previous few years, the original image can be recovered with more accuracy.
Final Deliverable of the Project Software SystemCore Industry ITOther IndustriesCore Technology Artificial Intelligence(AI)Other TechnologiesSustainable Development Goals Industry, Innovation and InfrastructureRequired Resources| Item Name | Type | No. of Units | Per Unit Cost (in Rs) | Total (in Rs) |
|---|---|---|---|---|
| Total in (Rs) | 80000 | |||
| GPU for Deep learning | Equipment | 1 | 55000 | 55000 |
| Random Access Memory | Equipment | 2 | 7500 | 15000 |
| Online Courses License | Miscellaneous | 1 | 8000 | 8000 |
| Printing | Miscellaneous | 1 | 500 | 500 |
| Possible Data Sources | Miscellaneous | 1 | 1500 | 1500 |