SNRGAN: The Semi Noise Reduction GAN for Image Denoising
Mehrshad Momen-Tayefeh, Mehrdad Momen-Tayefeh, Fatemeh Zahra Hasheminasab, S. AmirAli Gh. Ghahramani
Code Available — Be the first to reproduce this paper.
ReproduceCode
Abstract
Conventional noise reduction methods often fail to effectively handle high levels of noise, leading to artifacts and distortions. This paper proposes a Generative Adversarial Network (GAN) approach for noise reduction with low complexity. The proposed Semi Noise Reduction GAN (SNRGAN) effectively learns the underlying patterns of noise and generates denoised versions of noisy images, even with different noise levels. Training our model on three diverse datasets yielded admissible results, as evidenced by superior PSNR and NMSE scores. Furthermore, our model excelled in both subjective evaluations and objective metrics and its efficacy in handling elevated noise levels positions it as a promising solution for real-world applications.