Abstract
Image denoising, is a research problem where we aim to recover noise-free images from those that are contaminated with noise. It is also a very challenging problem for all researchers in the field of computer vision. There are various types of noise which differ on the basis of distribution and behaviour, e.g., salt-and-pepper noise, Poisson noise, Gaussian noise, etc. In this paper, we focus on eliminating Gaussian noise from contaminated images. Historically, many methods have been proposed for image denoising but recently, it has been observed that in most of the cases, Convolutional Neural Network (CNN) is used which outperforms the conventional methods. Here, we propose a CNN as the base model, to which we add other modules to improve the performance of the image denoising method. Our model uses an attention-guided CNN, called (ADNet), where we add median filter layers for restoring images contaminated by Gaussian noise. We apply median filters on all the feature channels of an image as well as increase the dilation rate of the dilated convolutions in the ADNet and so we see, in case of the convolutional layers, that the receptive field size is increased, which in turn aids image denoising. By quantitative analysis, we are able to show that our model performs significantly well when tested on the BSD500 and Set12 datasets.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have