基于生成对抗网络的人脸超分辨率重建算法

Face image super-resolution algorithm based on generative adversarial networks

为了能够生成纹理细节丰富的人脸图像,提出一种基于生成对抗网络的人脸超分辨率重建算法。设计一个由2条支路组成的生成器,并在网络中采用了更适合图像重建的残差单元结构;在模型优化过程中,提供一种将对抗损失与重建损失相融合的方法,有助于生成更加真实自然的人脸图像。实验结果表明,该重建算法能够得到细节突出的人脸图像,且FID分数高于目前领先的人脸超分辨率重建算法。

A novel face super-resolution algorithm based on generative adversarial networks is proposed to generate face images with more details in this paper. The generator is composed of two branches, which get the high-frequency and low-frequency components of the image, respectively. The residual unit structure is adopted and it is more suitable for image reconstruction in the network. In the process of model optimization, a fusion loss method is utilized, in which the loss of reconstruction and adversarial are weighted and fusioned to generate realistic and natural face images. The experimental results show that the face images with more details can be reconstructed by proposed method, and the FID score is higher than up-to-date face super resolution algorithms.