Dynamic Facial Expression Recognition based on Two-Stream-CNN with LBP-TOP

In this paper, we propose a dynamic facial expression recognition (FER) method based on two-stream architecture with both Spatial and Temporal Convolutional Neural Network (CNN) with LBP-TOP feature. The proposed system focuses on spatial feature with obvious expression frame and also focuses on the temporal information in all expressions sequence changed from non-expression frame. This two-stream architecture is validated in the field of action recognition in video by track the optical flow information on temporal part. In this paper, we utilized LBP-TOP feature which able to extract the spatial-temporal feature on facial expression change process, and is validated the effectiveness on this field. Our proposed method is evaluated using CK+. And the results are comparable to the state-of-the-art methods to prove the effectiveness of proposed architecture.