Computer Engineering and Applications ›› 2023, Vol. 59 ›› Issue (22): 166-173.DOI: 10.3778/j.issn.1002-8331.2207-0449

• Graphics and Image Processing • Previous Articles     Next Articles

Research on Age Synthesis of Adolescents Under Constraints of Kinship

LIN Xia, LI Jianwei   

  1. College of Physics and Information Engineering, Fuzhou University, Fuzhou 350116, China
  • Online:2023-11-15 Published:2023-11-15



  1. 福州大学 物理与信息工程学院,福州 350116

Abstract: Generation of realistic face images by age synthesis can effectively improve the accuracy of cross-age face verification, which is of great significance for finding the missing population. However, the immature skull complex of adolescents makes it very difficult to age synthesis. Therefore, an end-to-end age synthesis model for adolescents is proposed. The semantic information of the face is preserved through StyleGAN, the age channel is added to the face coding feature to realize the age conversion, the affinity feature matching module is introduced to guide the adolescents’ facial aging, and the affinity feature matching rate is added to the loss function in the training. The algorithm model can achieve smooth age synthesis and generate realistic face images while maintaining individual identity information, which not only improves the visual effect, but also shows that the accuracy of cross-age face verification reaches 95.3%, the identity recall rate reaches 92.7%, and the average age error of age synthesis decreases by 4 years, which is better than the existing algorithms.

Key words: age synthesis, StyleGAN, kinship verification, cross-age face verification

摘要: 年龄合成生成逼真的人脸图像可以有效地提高跨年龄人脸验证准确率,对寻找走失人口有着重要的意义,但是青少年颅骨复合体未发育完全,使得面向青少年的年龄合成十分困难。因此提出了一种面向青少年的端到端的年龄合成模型。通过StyleGAN保留人脸的语义信息,在人脸编码特征上添加年龄通道实现年龄的转化,引入亲缘特征匹配模块引导青少年的面部老化,将亲缘特征匹配率加入损失函数参与训练。该算法模型可以在保持个体身份信息的同时,实现平滑的年龄合成,生成逼真的人脸图像。该模型不仅提升了视觉效果,并且实验表明该模型跨年龄人脸验证准确率达到95.3%,身份召回率达到92.7%,年龄合成平均年龄误差减少4年,较现有算法有较好的提升。

关键词: 年龄合成, StyleGAN, 亲缘特征匹配, 跨年龄人脸验证