基于生成式对抗网络的图像修复研究进展

图像修复是图像处理的一个重要问题,目的是利用计算机视觉技术自动恢复退化图像中损坏或丢失的部分,被广泛应用于影视特技制作、图像编辑、数字化文物保护等领域。近几年,以生成式对抗网络(GAN)为代表的深度学习技术在计算机视觉和图像处理领域大获成功,基于GAN的图像修复逐渐成为主流,受到了广泛关注。针对图像修复的关键问题,文章对GAN和基于GAN的修复方法进行理论分析,首先整理分析了传统的基于人工特征的经典图像修复方法,其次总结了近年来基于GAN的代表性图像修复算法,并进行归纳分类,探讨了各类方法的特点和局限性。然后对图像修复模型常用的评价指标和公开数据集进行整理和分析,最后阐述了图像修复面临的挑战,对图像修复技术未来的发展方向进行展望。

Advances in Image Inpainting Based on Generative Adversarial Networks

Inpainting is an important problem in image processing, which aims to use computer vision technology to automatically restore damaged or lost parts in degraded images. It is widely used in film and television special effects production, image editing, digital heritage protection. In recent years, the deep learning technology represented by generative adversarial network(GAN) has achieved great success in the field of computer vision and image processing. And GANs based inpainting methods have gradually been widely concerned. Therefore, to cope with the key issues of image inpainting, we make a theoretical analysis of GAN and GAN based inpainting methods. We firstly summarize and classify the recent representative inpainting algorithms. In addition, we discuss the characteristics and limitations of these inpainting methods, and then organize and analyze the common evaluation indicators and public dataset. Finally, we describe the challenges of image restoration and prospect the future development direction of image restoration technology.