Method for Recognizing Facial Emotion Using A-V 2-Axis Emotion Model and Linear Regression
This technology converts human facial images into grayscale vectors, reduces dimensions via PCA, and numerically calculates the arousal and valence values of the A-V emotion model using linear regression analysis.
Existing emotion classification methods only recognize discrete, categorized basic emotions, which limits continuous human-robot interaction due to facial tracking failures or recognition errors.
By combining PCA with a linear regression model, this technology numerically estimates coordinate values in the A-V emotion space from continuous video frames in real time.
Key Features:
- Building a database that stores emotion-specific videos and the A-V value emotional states of facial expressions
- Preprocessing by vectorizing the eye and mouth regions of the emotion-specific videos stored in the database and calculating the mean vector
- Calculating independent variables through principal component analysis based on the distribution of the preprocessed emotion-specific videos
- Training a linear regression model with the calculated independent variables and applying input images to calculate emotional state values
