基于卷积神经网络的边缘保真图像去噪算法  

Edge-Fidelity Image Denoising Based on Convolutional Neural Network

作　　者：刘一畅 马伟 徐士彪[2] 张晓鹏[2] Liu Yichang;Ma Wei;Xu Shibiao;Zhang Xiaopeng(Faculty of Information Technology,Beijing University of Technology,Beijing 100124;National Laboratory of Pattern Recognition,Institute of Automation,Chinese Academy of Sciences,Beijing 100190)

机构地区：[1]北京工业大学信息学部,北京100124 [2]中国科学院自动化研究所模式识别国家重点实验室,北京100190

出　　处：《计算机辅助设计与图形学学报》2020年第11期1822-1831,共10页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(61771026,61971418,61671451);北京市科技计划(Z171100004417023).

摘　　要：现有图像去噪算法在去除噪声的同时,容易导致边缘过度光滑.为解决该问题,提出一种基于卷积神经网络的边缘保真去噪算法,它由基准去噪模块和基于多特征融合的边缘提取模块组成.首先,针对基准去噪模块所得结果,采用边缘提取网络提取边缘细节;进而,通过多层次边缘损失代价最小化,反向优化基准去噪网络去噪性能,引导其生成具有更多边缘细节信息的干净图像.在PyTorch环境下用常见的图像去噪数据集Set5,Set14,Kodak,McMaster,RNI15以及跨类型医学图像数据集上测试所提出算法,并与FFDNet等去噪算法进行对比.实验结果表明,所提出算法峰值信噪比值等指标均高于其他对比算法;在视觉效果上,所提出算法能够保留更多边缘细节和纹理特征,得到的去噪后图像更加清晰.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.

关 键 词：图像去噪 边缘提取 卷积神经网络 多特征融合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]