Research On Facial Expression Recognition Algorithm Based On Facial Key Points

Facial expression recognition refers to the emotional state of computer automatic perception of face image. It is a multiple disciplines cross subject related to muscle anatomy, image analysis, deep emotion psychology, human-computer interaction, automated, physiological and so on, which calls for basis of subjects above. At the same time, the research is of worth, and it will bring a great help to the development of the theory and the practical application of the real life. According to the functional task of the algorithm system, this paper divides the process of the research. The related algorithms are improved in this paper. The main work is as follows:1, Face detectionIn this paper, we use the face detection method based on Haar-like features and explore the principle and process of Haar classifier in detail. Experiments show the efficiency and accuracy of the human face detection method.2, Facial feature extraction(1) The innovation of expression representation. According to the facial expression coding system(FACS) theory, the facial action unit is used to describe the different types of facial expressions. The theory of psychology is directed at the nature of the occurrence of facial expressions. Describing expressions in this way, it is more reasonable and facilitates the establishment of a model. In addition, the facial action units can be described by the change of the facial key points, which can be effectively used to reflect the movement of facial muscles.(2) Improvement of facial key points location algorithm. In this paper, Facial key points location algorithm based on pair of points comparison feature and decision tree is proposed. On the basis of active shape model(ASM), the key points' local texture is extended from one-dimensional into two-dimensional and decision tree prediction model is used to replace the Mahalanobis distance minimization method. All the work above is to make improvements to the process of search new points in ASM. If so, possible influence of expression changes, color and other factors to gray feature could be avoided.3, Facial expression classificationDesign of a two level classifier: the first level classifier is used to extract the face feature Au, which makes the use of BP neural network method respectively to set up forecasting model for corresponding AU in three regions; the second level classifier is used to predict types of expression according to the AU which takes the use of BP neural network respectively for each expression to establish prediction model. AU is extracted with the prediction model and its output probability is used as the weight of the facial expression feature, which can effectively avoid the over-fitting problem of the second level classifier.The three part concatenated, construct a basic recognition system, namely the process of the algorithm based on facial key points proposed in the paper. The experimental results show that the proposed algorithm is efficient and robust.