Improving image processing and noise removal using generative deep learning techniques
TL;DR
In this study, a generative adversarial network (GAN) is introduced along with the U-Net architecture and has the capability to provide a powerful estimated fine image compared with a regular neural network approach.
Abstract
Denoising plays a crucial role in satellite and medical images. The classical methods struggle to find out the noisy values and might lead to some loss in image metadata’s essitals. In this study, a generative adversarial network (GAN) is introduced along with the U-Net architecture. GAN algorithm has the capability to provide a powerful estimated fine image compared with a regular neural network approach. The Modified National Institute of Standards and Technology (MNIST) dataset is injected with synthetic random noise. The proposed work demonstrated noise reduction capabilities and significant improvement in peak signal to noise ratio (PSNR), from 11.70 dB for noise image to 19.41 dB for the estimated image. The structural similarity index (SSIM) moved from 0.61 to 0.93, showing a big restoration with stable convergence for the image details. The GAN visual analysis based on deep learning showed effective output detail recovery.