Enhancing Facial Transformation Capabilities: Synthetic Child Facial Data Generation and Validation
K. Lahari, Chigarapalle Shoba Bindu, O. Roopa Devi · Advances in engineering research/Advances in Engineering Research · 2025
The rise of generative AI has significantly impacted content creation, facilitating the rapid generation of high-quality text, images, audio, and synthetic data.This study investigates the generation of synthetic facial data for children, enabling complex facial transformations such as expression changes, age progression, eye blinking, head pose variations, and modifications in skin and hair color under varying lighting conditions.Our dataset consists of over 300,000 unique samples sourced from ChildGAN and real-world images obtained from the Children's Vision Network.To evaluate the quality and distinctiveness of the generated facial features, we employ a variety of computer vision methodologies, including a CNN-based child gender classifier, face localization, facial landmark detection, identity similarity analysis using ArcFace, and assessments of eye detection and aspect ratio.The results demonstrate that highquality synthetic facial data can effectively mitigate the challenges of collecting extensive datasets from real children.Furthermore, this research aims to improve data augmentation techniques by utilizing diffusion models to generate child data samples that accurately represent ethnic diversity and various racial backgrounds.