Recognition of Emotions Based on Facial Expressions Using Bidirectional Long-Short-Term Memory and Machine Learning Techniques

Abstract

The most effective form of non-verbal communication that a person can use is found in their facial expressions. It is possible to gain insight into a person’s feelings by seeing the expressions on their face. This can also reveal information about the person’s level of discomfort, attentiveness, personality, social interaction, and physiological indications. The fact that facial expressions can be rather complicated and extremely varied depending on the individual being viewed presents a number of challenges for automatic facial expression analysis. In the work that has been proposed, some of the challenges that have been tackled include the extraction of facial features, the mapping of those features to highly discriminative spaces, the synthesis of emotions, the handling of multiple facial feature descriptors, and the detection of the intensity of Action Units. This is done with the intention of utilizing face components as opposed to holistic facial characteristics when conducting facial expression research. The approaches used to extract features need the extraction of face feature deformations from their normal states. These face feature deformations can be generated by a wide variety of mental states. The categorization of emotional qualities requires the utilization of both long-term and short-term memory in a bidirectional fashion (bi-LSTM). The approaches that are detailed in the work that is being presented have been applied to a standard database for the purpose of emotion recognition, pain intensity estimation, and facial action unit intensity detection so that the robustness of these methods may be demonstrated. The findings of the experiments show that the research work that was planned will be efficient as well as reliable.