Image denoising via deep network based on edge enhancement

Abstract

Existing methods for image denoising mainly focused on noise and visual artifacts too much but rarely mentioned the loss of edge information. In this paper, we propose a deep denoising network based on the residual learning and perceptual loss to generate high-quality denoised results. Inspired by the deep residual network, two new strategies are used to modify the original structure, which can improve the learning process by compressing the mapping range. At first, the high-frequency layer of noisy image is used as the input by removing background information. The secondly, a residual mapping is trained to predict the difference between clean and noisy images as output instead of the final denoised image. Furtherly improve the denoised result, a joint loss function is defined as the weighted sum of pixel-to-pixel Euclidean loss and perceptual loss. A well-trained convolutional neural network is connected to learn the semantic information we would like to measure in our perceptual loss. It encourages the train process to learn similar feature representations rather than match each low-level pixel, which can guide front denoising network to reconstruct more edges and details. As opposed to the standard denoising models that concentrate on one specific noise level, our single model can deal with the noise of unknown levels (i.e., blind denoising). The experiments show that our proposed network achieves superior performances and recovers majority of missing details from low-quality observations.