Ashwin, T. S.; Sanda, Nihar; Coursey, Austin; Gupta, Vaibhav; Biswas, Gautam. (2026).Ìý.ÌýProceedings of the ACM Symposium on Applied Computing, 87–94.Ìý
Artificial intelligence (AI) systems are increasingly being used to analyze students’ facial expressions to better understand engagement and emotions in classroom settings. However, using images of children raises important privacy and ethical concerns, making it essential to protect students’ identities without losing the emotional information needed for analysis. This study presents a new faceÌýde-identificationÌýmethod that replaces a person’s face with a realistic synthetic face while preserving the original facial expressions. The approach usesÌýStyleGAN, an AI model for generating realistic images, to select synthetic faces that match the emotional characteristics of the original face while removing identifying features. The method was evaluated using a real-world dataset of 40 middle school students and the publicly availableÌýDAiSEEdataset. Compared with conventional anonymization techniques, it achieved nearly 90% accuracy in protecting identity while maintaining high fidelity of facial expressions, with more than 92% agreement in manual evaluations and minimal loss of facial movement information. These findings suggest that the approach can better balance privacy protection with emotion recognition, supporting the development of privacy-preserving AI tools for classroom analytics and educational research.
