Fusing Edge-information in Image Denoising Based on CNN
Yichang Liu, Wei Ma · 2020
Due to the good performance, image denoising based on Convolutional Neural Network (CNN) has been widely studied. However, most of existing methods use a single neural network for image denoising. The denoised images occur smooth edges (missing details) at a high noise level. To address this problem, we propose a dual convolutional neural network for image denoising, which is termed as Fusing Edge-information in image Denoising based on CNN (FEDnets). It consists of two parallel network branches, which respectively get the denoised image and edge details in an end-to-end manner. In addition, the edges are fused with the denoised image to get a clearer and more detailed image. Experimental results show that FEDnets can be effectively applied to noise removal tasks and recover clearer images with more edge details and textures features.