Face Recognition Using the Improved SRGAN

• Published in: 2023 8th International Conference on Image, Vision and Computing (ICIVC) pp. 120 - 123
• Main Authors: Zhao, Yan, Yao, Chunlong, Li, Xu, Shen, Lan
• Format: Conference Proceeding
• Language: English
• Published: IEEE 27.07.2023
• Subjects: Adaptation models; Face recognition; Image quality; Image recognition; Robustness; self-attention; SRGAN; Superresolution; U-net; Visualization
• DOI: 10.1109/ICIVC58118.2023.10270679

Image super-resolution reconstruction is a pivotal issue in the field of computer vision, as it aims to generate high-resolution images from low-resolution ones. To tackle this task, an advanced deep learning model called SRGAN has been widely employed, which has proven effective in improving image quality. However, it still has certain limitations, especially when dealing with face images. In this paper, we propose an improved SRGAN model that includes two new improvements: Self-attention mechanism and U-net structure. To address the problem of face image super-resolution reconstruction, we trained the model on a large-scale face dataset. We introduced a self-attention mechanism to enhance the model's ability to extract detail information. The self-attention mechanism enables the model to adaptively focus on important regions in the input image, thereby better capturing the details of the image. Secondly, we added a U-net structure to the generator of SRGAN to enhance the model's perceptual ability and reduce the probability of generating visual artifacts. As a result, the performance and robustness of the model are effectively improved. We validated our proposed improved SRGAN model on multiple face datasets. Experimental results show that our model performs better in handling face images compared with the original SRGAN model. It can better preserve facial detail and texture information while reducing the generation of visual artifacts, thereby improving the accuracy of face recognition to some extent.