Emotion Recognition of Autistic Children from Facial Images Using Hybrid Model

Interpretation of an individual’s feelings can be done through analyzing their facial expressions and body movements. Emotion detection from facial expressions is a prominent research focus within the field of Artificial Intelligence. This approach is particularly significant in areas such as detecting autism in children, where precise diagnoses are important for personalized aid and support. However, it is challenging to predict the expression of kids with autism as one fails to read their feelings by seeing their reactions because they are unable to show reactions like normal human beings. Autism, sometimes known as autism spectrum disorder (ASD), encompasses a wide range of conditions characterized by challenges in communication, both verbal and nonverbal, repetitive behaviors, and social interaction skills. In this paper, we propose a new image dataset of autistic children named FERAC which has four classes (joy, anger, natural and fear) with the help of an existing dataset by Dr. Fatma M Talaat. In addition, we present a novel approach utilizing a hybrid model that integrates DenseNet121 and MobileNetV2 architectures to identify the emotions of autistic children from facial images. We developed and analyzed our hybrid models with four deep-learning models (CNN, VGG16, ResNet18, and IncerptionV3) and among them, our proposed hybrid model outperformed the other individual deep-learning models in terms of accuracy.