Işitme Engelli Çocuklarin Yüz Ifadeleri Tespiti

Facial emotions play a critical part in daily life and understanding and analyzing emotions are of great importance in terms of human-computer interaction. Although facial expressions between children and adults differ, many face recognition systems use models that are exclusively trained on adult data. Children with hearing impaired (HI) have hardship in social life due to differences in expressing emotions. Therefore, the facial expressions recognition of children with HI is a challenging problem. This study aims to develop a high performance emotion classification system for children with HI. For training Convolutional Neural Networks (CNN) from scratch, a large amount of data is required, but the children's dataset is limited. Firstly, fine-tuning the model with pre-trained CNN via transfer learning is discussed as a method for the classification of children's emotions. As a result of the experiments, the most accurate result was obtained by fine-tuning the dataset of children with 3 emotion labels with the trained-model of adults with 8 emotion labels.