基于渐进式GAN逆映射的人脸超分辨率重建  

Face Super-resolution Reconstruction Based on Progressive GAN Inversion

作　　者：孙红[1] 赵迎志 罗琦 袁巫凯 SUN Hong;ZHAO Yingzhi;LUO Qi;YUAN Wukai(School of Optical-Electrical and Computer Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学光电信息与计算机工程学院,上海200093

出　　处：《控制工程》2024年第9期1572-1580,共9页Control Engineering of China

基　　金：国家自然科学基金资助项目(61472256,61170277,61703277)。

摘　　要：为了缓解生成对抗网络(generative adversarial network,GAN)自身存在的训练不稳定问题,增强图像超分辨率重建的效果,提出一种基于GAN逆映射的图像超分辨率重建算法。通过渐进式网络增加模型的稳定性,通过逆映射金字塔充分学习低分辨人脸图像的语义特征,实现准确的隐空间转换,通过半空间特征调制进行图像增强。对所提算法进行实验验证,该算法在Celeb A数据集上重建8倍后的图像峰值信噪比为27.18 dB,相比超分辨率生成对抗网络(super-resolution generative adversarial network,SRGAN),提高了2.44 dB,模型的收敛速度也更快。实验结果表明,通过GAN逆映射和渐进的方式进行图像超分辨率重建具有较好的效果和更高的稳定性。In order to alleviate the training instability of GAN and enhance the effect of image super-resolution reconstruction,an image super-resolution reconstruction algorithm based on GAN inversion is proposed.Taking the progressive network as the overall architecture to increase the stability of the model,fully learn the features of lower resolution images through the inverse mapping pyramid,carry out accurate hidden space conversion,and enhance the images through half space feature modulation.The experimental results show that the PSNR of the algorithm is 27.18 dB,which is 2.44 dB higher than SRGAN,and the convergence speed of the model is faster.The experimental results show that the image super-resolution reconstruction using GAN inversion and progressive method has better effect and higher stability.

关 键 词：半空间特征调制 逆映射金字塔 渐进式网络 超分辨率重建 生成对抗网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]