Edge-Fidelity Image Denoising Based on Convolutional Neural Network

Image denoising is a hot research topic in the fields of image processing and computer vision.It aims to estimate a latent clean image from a noisy one.Existing denoising algorithms generally result over smooth edges while removing noises from the image.To eliminate this problem,we propose an edge-fidelity denoising deep model based on convolutional neural network(CNN).It is composed of a basic denoising module and a multi-feature fusion-based edge detection module.We use the edge detection module to extract multi-level edges from the output of the denoising module.Then,we guide the denoising module to generate edge-fidelity results by minimizing multi-level edge losses,in an end-to-end manner.Experimental results show that the proposed network obtains higher PSNR values than state-of-the-art methods on multiple datasets,including Set5,Set14,Kodak,McMaster,RNI15 and a dataset composed of medical images,and recovers visually clearer images with more edge details and textures.