Research for Face Image Super-Resolution Reconstruction Based on Wavelet Transform and SRGAN

• Published in: 2021 IEEE 5th Advanced Information Technology, Electronic and Automation Control Conference (IAEAC) Vol. 5; pp. 448 - 451
• Main Authors: Cao, Minjie, Liu, Zhe, Huang, Xueting, Shen, Zhuoxuan
• Format: Conference Proceeding
• Language: English
• Published: IEEE 12.03.2021
• Subjects: Face recognition; Feature extraction; Generative adversarial networks; Image reconstruction; Reconstruction algorithms; super-resolution; super-resolution generative adversarial network(SRGAN); Superresolution; Wavelet coefficients; wavelet transform
• ISSN: 2689-6621
• DOI: 10.1109/IAEAC50856.2021.9390748

Super-resolution face image is the basis of high detection rate in face recognition. In order to meet the requirements of super-resolution image in face recognition, aiming at the problem of texture loss of super-resolution image under high-frequency features, a face image reconstruction method based on wavelet transform and super-resolution generative adversarial network (SRGAN) is proposed to reduce the impact of low-resolution image caused by imaging hardware, network bandwidth and sampling environment on face recognition accuracy. Firstly, the wavelet transform algorithm is used to preprocess the low-resolution face image to extract the detailed texture features of the face image under different frequencies. Then, GAN is used to learn the prior knowledge of wavelet coefficients, and the identity preserving constraint is applied to the output image, and the perceptual loss function of the fusion wavelet coefficients is realized. Finally, the deep learning model based on SRGAN is used to obtain high-resolution face images. Experimental results show that the method can achieve super-resolution restoration of low-resolution face images and meet the requirements of face recognition accuracy.