基于生成对抗网络的面部图像修复算法研究
Research on Image Restoration Algorithm Based on Generative Adversarial Networks
张研 1刘晓群1
作者信息
- 1. 河北建筑工程学院,河北张家口 075000
- 折叠
摘要
为了解决目前基于生成对抗网络的图像修复算法在修复大范围面部图像时修复效果不好,特征提取不充分的问题,提出了基于注意力机制的生成对抗网络修复模型.引入自注意力机制模块来感受图像全局特征用来生成图像缺失区域,能够更好地修复图像的大范围缺失.同时改进了判别器,引入感知损失,提高了修复图像的结构相似度.在CelebA数据集上的实验结果表明,该算法在各项评价指标上均优于现有主流算法,其PSNR损失提高了 0.81％～1.94％,SSIM 提高了 1.06％～2.49％,MSE 降低了 0.12％～0.35％.
Abstract
In order to solve the problem that the current image inpainting algorithm based on generative adversarial network has insufficient feature extraction when repairing large-scale face images,and can only use small-scale feature information.In this paper,a generative adversarial network repair model based on attention mechanism is proposed.The self-attention mechanism module is intro-duced to feel the global features of the image to generate the missing area of the image,which can better repair the large-scale missing area of the image.At the same time,the discriminator is im-proved,the perceptual loss is introduced,and the structural similarity of the restored image is im-proved.Experimental results on the CelebA dataset show that the proposed algorithm is superior to the existing mainstream algorithms in various evaluation indexes,The PSNR loss increased by 0.81％～1.94％,the SSIM increased by 1.06％～2.49％,and the MSE decreased by 0.12％～0.35％.
关键词
生成对抗网络/图像修复/注意力机制/感知损失Key words
generative adversarial networks/image inpainting/Attention mechanisms/Perceived loss引用本文复制引用
出版年
2024
NETL
NSTL
万方数据