Autoencoder For Denoising Images

• As input of the encoder we will give image with added noise (usually noise from Gaussian Distribution)
• And the output will be image without noise
• By training, autoencoder based on this input & output, it will learn to generate important latent features from noisy images
• And use that features to generate denoisy image
• Then this model can be used to denoise any noisy image
• This idea can be extended to any modal, i.e. text