Assistive facial expression recognition for children with autism using re-enactment

Accurately predicting and interpreting facial expressions remains a significant challenge, particularly for individuals with social impairments or Autism Spectrum Disorder diagnoses. This study proposes a novel approach to enhance Facial Expression Recognition through the generation of augmented facial expression images. The method integrates a facial re-enactment model with a dedicated Facial Expression Recognition classifier to generate enhanced images aimed at improving expression clarity and interpretability. A clinical study involving 48 participants, including individuals with Autism Spectrum Disorder, was conducted to evaluate the effectiveness of the enhanced images in supporting facial expression recognition. Additionally, a custom classification model based on the EfficientNet-v2 architecture was trained on the AffectNet Database, achieving 67.83% accuracy for seven expression categories and 64.65% accuracy for eight. Participants showed improved performance when recognising expressions in enhanced images, with particularly positive results observed among those with Autism Spectrum Disorder. These findings demonstrate the potential of automated enhancement techniques to support assistive Facial Expression Recognition applications and highlight a scalable direction for future interventions targeting individuals with social or cognitive impairments. • Enhanced recognition of facial expressions for individuals with autism spectrum disorder. • Novel re-enactment model generates augmented facial expressions for improved interpretation. • EfficientNet-v2 classifier achieves significant accuracy gains in emotion recognition. • Clinical study validates the benefits of enhanced facial expression images. • Practical applications for assistive technology and social interaction support.