Estimation of Facial Emotion Based on Landmark Points by Applying Artificial Intelligence and Machine Learning

In the science of emotion processing, it is hard to figure out how to predict an emotion from naturalistic facial expressions in a multidimensional space. While many studies and works have been done to develop methods for making accurate predictions, relatively little has been done to put the uniqueness of the human face to use in reading and responding to people's expressions of emotion. This work proposes an emotional image retrieval strategy for multimodal emotion detection. This strategy is based on the fact that face features are always there and that the human face has regional redundancy that is independent of face features. Our method uses landmark points to get rid of unnecessary information about facial expressions and focus on the most attractive ones in each frame. It's possible that by making use of the landmarks, we'll be able to predict facial emotions more precisely. This is especially helpful for long-term emotion prediction since it allows us to recover time and space representations by introducing an emotional face. The preliminary results based on a benchmark dataset show that the proposed method exceeds its competitors in terms of prediction accuracy.