Facial Expression Recognition for Children: Can Existing Methods Tuned for Adults Be Adopted for Children?

Child facial expression recognition plays an important role in child-machine interactions. Recognition methods that were tuned for adults have been used for children in many studies without evaluating the applicability of these methods on children. This paper investigates this problem using a Support Vector Machine classification-based recognition algorithm, which is one of the most widely applied methods. We examined: (1) the difference in facial expressions between children and adults and (2) whether the classifiers trained on one group work for the other. Results show that the classifiers trained on child data were more accurate when tested on child data than adult data and the classifiers trained by adult data were more accurate when tested on adult data than child data. When the training and testing data were from the same group, the classifiers generally performed better for adults than children. Implications and future works are discussed with the results.