FACIAL EXPRESSION RECOGNITION WITH DEEP AGE

This paper presents a novel deep learning framework for facial expression recognition (FER). Our framework is derived from Convolutional Neural Networks (CNNs), adopting outline, texture, and angle raw data to generate 3 different convolutional feature maps for deep learning. In so doing, the proposed method is cable of robustly classifying expressions, by emphasizing the facial deformation, muscle movement, and outline feature descriptions. Therefore, our method makes the facial expression task not only more tractable, but also more reliable in the case of expression images, which leads to an overall improvement of classification performance compared to conventional methods. The proposed method is valid for the Static Facial Expression Recognition (SFEW) database, improving the baseline results by 6.98%.