一种生成对抗网络用于图像修复的方法

近年来基于深度学习的图像修复方法相比于传统方法，表现出明显优势，前者能更好的生成视觉上合理的图像结构和纹理.但现有的标准卷积神经网络方法，通常会造成颜色差异过大和图像纹理缺失与失真的问题.本文提出了一种新型图像修复深度网络模型，该模型由两个相互独立的生成对抗式网络模块组成.其中，图像修复网络模块旨在解决图像缺失区域的修复问题，其生成器基于部分卷积网络；图像优化网络模块旨在解决修复后图像存在局部色差的问题，其生成器基于深度残差网络.通过两个网络模块的协同作用，图像的视觉效果与图像质量得到提高.与其他先进方法进行定性和定量比较的实验结果表明，本文提出的方法在图像修复质量上表现更好.

Image Inpainting Using Generative Adversarial Networks

In recent years,deep learning based methods have shown preferable results for the task of inpainting corrupted images.However,the existing standard convolutional neural network approaches often cause problems with excessive color discrepancy,image texture loss and distortion.A deep network based image inpainting model is proposed in this paper,consisting of two generative adversarial network modules.One of the modules is used to inpaint the missing area of the image,where the generator is constituted with partial convolutions.The other module is the image optimization network,which is applied to solve the problem of local chromatic aberration after image restoration,and in which the generator is originated from the depth residual network.These two modules cooperated to improve the visual effect and image quality of inpainted images.Using MOS,SSIM and PSRN as the evaluation criteria,the experimental results of qualitative and quantitative comparisons with other state-of-the-art methods have shown that the proposed model performed better.