Facial Expression Recognition In Video Sequences

Facial expression is an important carrier of transferring body behavior information and emotion in daily communication between people. Depth study of facial expression recognition has evident significance for a better understanding of human psychology.Besides, with the continuous progress of science and technology and the continuous development of society, more and more areas need to apply the facial expression recognition technology in the video sequence. Since the position and the tilt angle of the faces in video sequences are likely to change at any time, therefore the facial expression recognition system has higher requirements on processing ability of identification algorithm and rotational invariance of the extracted features. So in this paper, we conduct a series of significant research and exploration for the facial expression recognition technology in the video sequence on the basis of previous work, which includes the following aspects.First, based on the use of Haar feature and AdaBoost algorithm we achieve the face detection function which is very important for the whole system. Moreover, we combine the skin color information with face detection method to reduce erroneous recognition results that making detection algorithm get the face position which acquired from the image more accurately and efficiently. It plays an very important role in the following facial expression recognition step and helps establishing a good foundation for the final result.Then, we use the Local Binary Pattern Histogram Fourier features(LBP-HF) to extract recognition vector from the face areas which achieved from the previous face detection step and we use the Support Vector Machine to conduct the facial expression classification. LBP-HF feature is rotation invariant and has a better texture recognition ability. In addition, LBP-HF feature could be calculated very simply and efficiently.Therefore it could satisfy the real-time requirement. We conduct the comparative experiments on multiple common facial expression datasets and the experimental results demonstrate that the facial expression recognition method based on LBP-HF feature has a rather good effectiveness and application value.Finally, we decode the video files to extract the image by using FFmpeg library which is the leading multimedia framework. Thus we achieve the real-time expression recognition system in video sequences by combining the face detection algorithm and the facial expression recognition method.