Emotion recognition using fiducial points via deep learning
Emotion awareness is critical because of the function that emotions play in our daily lives. As a result, automatic emotion recognition aims to provide a machine with the human ability to interpret and comprehend a person's emotional state in order to predict his intents based on his facial expression. In this research, a new method for improving the accuracy of emotion recognition from facial expression is proposed, which is based solely on input attributes deduced from fiducial points. First, 1200 dynamic features representing the percentage of euclidean distances between facial fiducial points in the first frame and facial fiducial points in the last frame are extracted from image sequences. Second, just the most relevant features are chosen using active learning method. Finally, to categorise facial expression input into emotion, the selected features are provided to a ResNet classifier.