Research On Related Issues Of Facial Features Extraction Algorithm In AU Recognition

With the rapid development of information technology and artificial intelligencetechnology,the field of intelligent recognition about technology the human face hasbeen rapid development.Facial action unit automatic identification can be applied tointelligent robot research, can also be used in psychological research, health care,public safety, etc., so in recent years at home and abroad it has become a researchhotspot.The facial feature extraction is the premise and foundation of AU recognition,so the efficient facial feature extraction algorithm is of great significance.This articlemain research content includes:First,we aiming at the problem that the steepest descent method in the fittingprocess of Active Appearance model may not convergence because of the it cannotdetermine the step length,this article use Marguardt method to optimize.This methodcan better convergence, and a certain extent reduce the fitting error.Second,aiming atthe problem that it fails to locate because of the iteration of the fitting can notconverge to the correct location when the initial position is far away from thedestination,a AAM facial feature point tracking method using prediction of strongtracking filter (STF-AAM) was proposed. This method can find the fitting initialposition of each frame of video sequences and achieve a more accurate and morerapid tracking result. The experimental results show that the proposed methodperforms better than traditional method in the tracking speed along with the fittingaccuracy case.Third,facial feature extraction system was designed andimplemented.Facial feature of Xinjiang Uygur-Kazak facial expression databasewhich is built by our team are extracted and verified the effectiveness of extractedfeatures to identify the AU by two BP neural networks.Fourth,aiming at the problemthat we difficultly score some combination of AU,this article analysis and summarizessome methods that we score of combination of AU.Uygur-Kazak facial expressiondatabase is introduced, and they are scored.