Copyright © 2026 Authors retain the copyright of this article. This article is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

@article{208397,
        author = {BARUN BISWAS},
        title = {Document Image Denoising using Autoencoder with GAN},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {13},
        number = {4},
        pages = {1765-1782},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=208397},
        abstract = {Cleaning noise from document images is an important first step to make text easier to read and improve the accuracy of OCR systems, digital archiving, and automatic document analysis. However, using traditional noise removal methods (Gaussian filtering, median filtering, and wavelet techniques) often blurs small text details and information of the document is lost. To overcome these problems, we introduce a CNN-GAN based method for document image denoising. The model recognize different complex noise patterns and produce clean images without hampering important text and structure. Our proposed CNN model takes noisy document images as input and produces clear, noise-free images using several convolution, activation, and up sampling layers along with the function of generator and discriminator of GAN. The network is trained on a dataset of noisy document images like tobacco800 and USDFUN. L1 loss (Mean Absolute Error) is used for training and peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM) is used to calculate the performance of the network model. Experiments show that the CNN-GAN based method performs better than traditional denoising techniques and some available denoising methods which use simple CNN architecture removing noise more effectively without hampering the text, edges, and fine details. This study shows that deep learning methods can greatly improve the quality of document images and help make OCR more accurate.},
        keywords = {Document Image Denoising, Generative Adversarial Network (GAN), Autoencoder, Convolutional Neural Network (CNN), Reconstruction Loss, Structural Similarity Index (SSIM), tic Noise Generation},
        month = {September},
        }