An Experimental Study on Denoising the Images with Autoencoders

Noisy data is still one of the most common issues in modern data transmission. We can solve this by using deep learning with an autoencoder, a feature extraction method that can reduce noise. In the current study, we examine the ability of an Autoencoder to minimize noise on various structures and amounts of additive noise implemented with multiple optimizers. The experiment is run on a Convolutional Denoising Autoencoder (CDAE), primarily for image data. The experiment is run with additive noise with a noise factor of 0.03, such as Gaussian, Salt and Pepper, and Speckle. The proposed model outperformed image quality measures PSNR, MSE, SSIM, and RMSE. The reconstructed images after denoising are identical to the original image data.