End to End Deep Neural Network: An approach to clean noisy documents

Digitalization is required for the industry's evolution in each sub-sector. While the rate of digitalization has had an influence, traditional methods of document storage have been supplanted by digital techniques such as databases, softcopies, and so on. However, certain industries continue to use the conventional technique of storing data, such as hard copies and scanned digital documents. There would be a certainty of the presence of numerous factors, such as noise, background noise, obscured and blurred text, and watermarks, while collecting digital photos from these hard copies and scanned documents. The quality of the acquired image degrades as a result of these factors. Because of the variables listed above, the digital version, i.e., photographs of the physical document, has declined in quality. This article proposes a solution to this problem. The proposed model loosely follows the Encoder-Decoder structure. The model is trained on 2000 images and managed to achieve 66.33% accuracy on the validation set.