Download PDFOpen PDF in browser

Image Restoration using Deep Learning Techniques: a Dataset Free Approach

EasyChair Preprint 10491

6 pagesDate: July 2, 2023

Abstract

Image restoration is a primary computer vision task which involves returning damaged images to their original state. Recently, deep learning has revolutionized the field of image restoration by delivering cutting-edge outcomes on a variety of restoration jobs. This study examines the use of deep learning techniques for image noise reduction and enhancing image resolution. Specifically, we investigate the use of deep neural networks trained on a single noisy image to restore a clean image. The architecture of the ConvNet, a popular neural network model for image denoising and enhancing image resolution, also discussed the latest advances such as residual connections and UNet with skip connections. PSNR and MSE among diverse pictures are used to evaluate the performance of our technique on various image denoising and resolution enhancement tasks. The results demonstrate that deep learning-based image denoising and high resolution can achieve high-quality results and they have the ability that can be used in numerous applications without being trained on a dataset. The traditional CNN is used to denoise the image using reliable functions in a sequential model. To obtain picture super resolution, a randomly initialized ConvNet with untrained convolutional layers are used where the model is not trained on a dataset but rather only one image.

Keyphrases: CNN, ConvNet, Image denoising, SISR, UNet, deep learning, image restoration, noise, single image super-resolution

BibTeX entry
BibTeX does not have the right entry for preprints. This is a hack for producing the correct reference:
@booklet{EasyChair:10491,
  author    = {Abhinav Bhogavalli and Harsha Vardhan Chintamaneni and Purandhar Tatarao Adigarla and Suma Kamalesh Gandhimathi},
  title     = {Image Restoration using Deep Learning Techniques: a Dataset Free Approach},
  howpublished = {EasyChair Preprint 10491},
  year      = {EasyChair, 2023}}
Download PDFOpen PDF in browser