A Clean to Noisy Image Generation Scheme Using Generative Adversarial Network

EasyChair Preprint no. 9338

Abstract

For training, current deep learning real denoising algorithms need a lot of noisy, clean image pairings. Nevertheless, it is an extremely expensive and time-consuming process to capture a true noisy-clean dataset. To address this issue we looks towards creating realistic noisy visuals. We propose a generative adversarial network (GAN) based noise generation model which utilizes a pre-trained image denoiser to construct the fake and real noisy images into a nearly noise-free solution space. Utilizing this denoiser we have developed a network to generate realistic looking noisy images.

Keyphrases: Generative Adversarial Network, Image denoising, Noisy image generation

@Booklet{EasyChair:9338,
  author = {Sadat Hossain and Bumshik Lee},
  title = {A Clean to Noisy Image Generation Scheme Using Generative Adversarial Network},
  howpublished = {EasyChair Preprint no. 9338},

  year = {EasyChair, 2022}}