Video-based Facial Expression Recognition Using DensNet۱۲۱ andLSTM

Facial expression is a non-verbal communication that emerges from humans' innerfeelings and allows us to be aware of people's emotions without verbal communication. Facialexpression recognition techniques are required in various today’s artificial intelligence-basedtechnologies such as automobile driving, human-machine interface, and market assistant. This studypresents an approach to classifying facial expressions using a Convolutional Neural Network (CNN)with Long Short-Term Memory (LSTM) network. Particularly, CNN (DenseNet۱۲۱) is used toextract the features of the incoming video frames. The pre-trained network DenseNet۱۲۱ is employedto surpass the need for a high number of training samples. LSTM network is employed for attainingthe temporal patterns of the sequence of images to avoid gradient explosion. Then, the facialexpressions are classified into seven different types of emotions (Anger, Disgust, Neutral, Sadness,Fear, Happiness, and Surprise). The results show that the proposed method reaches higherclassification accuracy (۷۲.۷۶%) than three other competing methods on the BAUM-۱ dataset.

کلیدواژه ها:

Facial Expression ، Long Short-Term Memory ، Transfer Learning ، Convolutional Neural Network