Nmf Facial Expression Recognition

Non-negative Matrix Factorization (NMF) is a new method issued by international workers. Compared to other matrix factorization, the great advantage of NMF is to bring the non-negative restriction to the process of matrix factorization, which can show the original data and reflect its part features.Facial expression recognition is a scientific topic developed by recent decades. It apply the computer to analyze the given facial expression and its changes, then judge the inner emotion and thought activity. So communication between individual and machine becomes more physical and intellectualized. It is being widely applied in many fields, such as Psychological research, Image understanding, Synthesis facial animation, Video retrieval, Robot technology and new style human-computer interaction based on facial expression.This thesis adopt the NMF into the operation of human facial expression recognition.The main contents includes:(1) This thesis studies the basic theory of NMF algorithm and discusses the way to extract the facial expression feature by NMF. Non-negative restriction is involved in the process of expression feature extract, which means to reconstruct the expression image in a non-subtractive way, to ensure the process of forming unity from parts.(2)The NMF algorithm was employed to extract every expression feature in JAFFE and CED-WYU (1.0) ficial database.And then this thesis studies how to select the number of basic image. The method of NMF is compared with principal component analysis (PCA) in extracting the facial expression feature. Experimental results show that the NMF algorithm can more effectively reflect the characteristics of Facial Expression(3) The NMF algorithm was emplyed to extract expression feature of facial expression database, then selected support vector machine as a classifier for Facial Expression Recognition.The relation of basici image number and system recognition rate was discussed.(4) The NMF algorithm and other feature extract algorithm such as linear discriminant analysis (LDA) were emplyed to extract expression feature of facial expression database.By experiment on facial expression database, NMF algorithm achieve a higher recognition rate than other feature extract algorithm in facial expression recognition. So NMF algorithm to extract the facial expression feature is feasible and effective.