Prediction of face age progression with generative adversarial networks

Area of Science:

• Computer Vision
• Artificial Intelligence
• Image Processing

Background:

• Face age progression is crucial for security and identification systems.
• Existing methods suffer from unnatural facial attribute modifications and visual artifacts.
• Accurate generation of aged faces remains a significant research challenge.

Purpose of the Study:

• To develop a realistic face aging method addressing limitations of current approaches.
• To enhance the accuracy and visual quality of synthesized aged faces.
• To explore practical applications of super-resolution face aging.

Main Methods:

• Utilized Attention Generative Adversarial Networks (AttentionGAN) with separate subnets for attention and content masks.
• Applied regex filtering to isolate synthesized faces from AttentionGAN output.
• Employed image sharpening and edge enhancement for high-quality input to Super-Resolution Generative Adversarial Networks (SRGAN).

Main Results:

• Generated high-quality, super-resolution face-aged images with detailed information.
• Achieved a synthesized face aged image error rate of 0.001%.
• Experimental results validated on five public datasets (UTKFace, CACD, FGNET, IMDB-WIKI, CelebA).

Conclusions:

• The proposed AttentionGAN and SRGAN-based method effectively generates realistic and high-quality aged faces.
• The technique offers improvements over prior face aging methods, demonstrated through quantitative and qualitative evaluations.
• The work highlights the practical utility of super-resolution face aging using GANs in various applications.