METRIC LEARNING BASED SEMI-SUPERVISED LEARNING ALGORITHM FOR REAL-TIME EMOTION RECOGNITION APPLICATIONS

摘要

A semi-supervised learning algorithm based on metrology learning for a real-time emotion recognition application is disclosed. A semi-map learning method for emotion recognition performed by an emotion recognition system according to an embodiment includes: detecting a face region from each frame of video data as video data is received; performing pseudo-label-based semi-supervised learning on the detected face region using a learning model in which two or more learning networks are combined; and estimating emotion information on the detected face region as a learning result obtained through the performed semi-supervised learning.