A super resolution frontal face generation model based on 3DDFA and CBAM

Abstract

Pose and low resolution seriously affect the synthesis of high-quality frontal face images. With the development of deep learning, a large number of models based on the deep neural network are used to solve the problem of face pose and image super-resolution. However, the synthesis of the high-resolution frontal face is still a problem that has not been fully studied. Therefore, in this paper, we propose a method to realize image super-resolution and face frontal generation simultaneously. Specifically, we propose a frontal face model FFSR_GAN used to generate super-resolution. This model mainly solves the problem of low resolution and large face pose. There are two main improvements: 1) Aiming at the problem of artifacts in the image generated by the face frontal generation module, the face frontal generation module is designed based on 3DDFA and CBAM; 2) Aiming at the problem of low resolution in frontal face generation, a face super-resolution module is carefully designed, which is used for super-resolution of the generated frontal face. The method proposed in this paper solves the problem of face pose and super-resolution for the first time and improves the recognition accuracy of low-resolution and face images with larger posture. The experimental results on the existing public dataset prove the advantages of the FFSR_GAN model.