Research On Methods For Facial Expression Intensity Measurement Based On Video

As the development of affective computing and intelligent Human-Computer Interface (HCI), machines are required to have the ability of apperceiving and understanding man's feeling and emotion. Under this circumstance, facial expression analysis came up as an important task and it has been widely studied in the recent years. Now, there're many methods of facial expression recognition, but research on expression intensity measurement hasn't been developed. Measurement of facial expression intensity is helpful to understand man's emotion state and emotion intensity, it is the development trend of affective computer.This paper first summarizes the current research state of facial expression analysis, then the method of facial expression intensity measurement is studied and a expression intensity measurement system is constructed. The main work is as follows:1.An improved method of facial feature point tracking is proposed in this paper. L-K optical flow algorithm is not very suitable in facial feature point tracking. A method of revising tracking on the basis of L-K optical flow is used. By using the geometric position relationship among facial feature points as constraint, feature points that were inaccurately tracked are tracked again as a revision. The veracity of facial feature points tracking is improved.2.In the section of facial feature dimensionality reduction, a nonlinear dimensionality reduction method-Isometric feature mapping (Isomap) is used to extract the 1-d manifold which represents the expression intensity from the high dimensional facial feature point trajectories. The expression intensity spectrum of the training sequences is constructed automatically.3.Support Vector Machine is used to build the expression intensity model for specific expression of specific subject and five intensity grades are introduced. Expression intensity value of traing frames are extracted from Isomap, and the frame is divided into one of the five grades based on the intensity value. Training facial feature point trajectories and corresponding expression intensity grades are used as learning samples of SVM, and expression intensity model is built to classify the expression frames in the test set.4.Using the above methods, a facial expression intensity measurement system is constructed and happy expression of a specific subject is measured. The experiment results show that the system is of validity in expression intensity measurement.