Journal of Frontiers of Computer Science and Technology ›› 2019, Vol. 13 ›› Issue (8): 1402-1410.DOI: 10.3778/j.issn.1673-9418.1811025

Previous Articles     Next Articles

Application of Generative Adversarial Networks in Image Completion

SHI Cheng, PAN Bin, GUO Xiaoming, LI Qinqin, ZHANG Luyue, ZHONG Fan   

  1. 1.School of Computer and Communication Engineering, Liaoning Shihua University, Fushun, Liaoning 113001, China
    2.College of Sciences, Liaoning Shihua University, Fushun, Liaoning 113001, China
    3.School of Computer Science and Technology, Shandong University, Qingdao, Shandong 266237, China
  • Online:2019-08-01 Published:2019-08-07

生成式对抗网络在图像补全中的应用

时澄潘斌郭小明李芹芹张露月钟凡   

  1. 1.辽宁石油化工大学 计算机与通信工程学院,辽宁 抚顺 113001
    2.辽宁石油化工大学 理学院,辽宁 抚顺 113001
    3.山东大学 计算机科学与技术学院,山东 青岛 266237

Abstract: Image completion is an important research direction in the field of digital image processing and has broad application prospects. This paper proposes an image completion method based on generative adversarial networks (GAN). The generative adversarial networks model consists of two parts: the generator model and the discriminator model, all of which are implemented by convolutional neural network (CNN). Firstly, the missing region of the image is complemented by the generator model. Then, the complemented image is discriminated by the discriminator model. In order to enhance the processing power of image texture details, Markov random field (MRF) and mean square   error (MSE) are used as the loss function to train the generator model. The experimental results show that the image completion method based on the generative adversarial networks has better completion effect than other existing methods.

Key words: image completion, generative adversarial networks (GAN), convolutional neural network (CNN), Markov random field (MRF), mean square error (MSE)

摘要: 图像补全是数字图像处理领域的重要研究方向,具有广阔的应用前景。提出了一种基于生成式对抗网络(GAN)的图像补全方法。生成式对抗网络模型由生成器模型和判别器模型两部分构成,通过采用卷积神经网络(CNN)实现。首先,通过生成器模型对图像的缺失区域进行补全;然后,利用判别器模型对图像的补全效果进行判别。采用马尔科夫随机场(MRF)与均方误差(MSE)相结合的损失函数训练生成器模型,加强对图像纹理细节的处理能力。实验结果证明,基于生成式对抗网络的图像补全方法,相较于其他现有的方法,具有更好的补全效果。

关键词: 图像补全, 生成式对抗网络, 卷积神经网络, 马尔科夫随机场, 均方误差