Noise reduction in images using autoencoders

Abstract

Pure signals can ideally exist only on paper. However, with the availability of a considerable number of state-of-the-art techniques that claim to denoise a given signal only up to an extent, it is at the same time necessary that such techniques must be compatible with a large number of devices. This paper discusses an approach to reduce the noise present in images through Image Processing and Deep Learning algorithms by implementing autoencoders. Using an autoencoder for removing noise is one such approach where the main focus is to retain the originality as autoencoders follow the backpropagation process, rather than the traditional techniques, where the output signal is ultimately the unoriginal version of the input signal. The technique discussed in the paper is interconvertible, i.e., could be used for any signal, is reliable, efficient, and compatible with a larger pool of devices.