Research On Upper Facial Action Units State Recognition

Facial action unit recognition is to analysis the states of different single facial action units and their combinations by computer. With facial action unit recognition, man can ascertain the subject's specific inborn emotion, achieving smarter and more natural interaction between human and computer. The study of facial action unit recognition has found important realistic values for enhancing computer intelligence, developing new human-computer environment, and promoting psychology research, which will create great economical and social benefit.In this thesis, we firstly discuss the background and then analyze the main facial action unit recognition algorithms presented, emphasizing on wavelet transformation, Independent Components Analysis (ICA), Moment invariants and Optical Flow Models (OFM). Then for the main processes in the facial action unit recognition, that is, facial image preprocessing, action unit location and segmentation, action unit feature extraction and recognition, we present some novel algorithms. They are described as follows(1) The algorithm of upper facial action unit location and segmentation based on Harris corner detection is proposed. On the base of face region image detectation, the image is preprocessed. Then every upper facial action unit is located by using the algorithm based on Harris comer detection and is divided from face image. This algorithm can divide different upper facial action unit exactly from face image.(2) The algorithm of upper facial action unit feature extraction based on KPCA is presented. After upper facial action unit location and segmentation, we present the facial action unit feature extraction algorithm based on KPCA. In KPCA algorithm design, according to the characteristics of upper facial action unit image, we choose and design corresponding kernel function and form the improved algorithm to extract KPCA features reasonably. This algorithm not only can mask different human features and illumination variation, recognize the action units that is subject independent, which makes the system more reliable and robust, but also can map the image onto the feature space, which reduce dimension of image matrix hugely and reduce computational cost(3) The classifier based on the improved SVM is used to recognize upper facial action unit state. Through analyzing and experimenting on the characteristic of KPCA features, the relative kernel function and chastisement factor are chosen and designed. Finally, the upper facial action unit state recognition algorithm based on SVM is built. The algorithm can obtain superior recognition rate and has fast speed.In the end of this thesis, we design and develop a prototype system of upper facial action unit feature extraction and recognition based on object-orient methods and using corresponding library functions in VC++ and Matlab. The system running proves that above algorithms are correct and effective.