|
Facial Expression Recognition (Face Expression Recognition, FER) is the use of computer technology for feature extraction of facial expression information to be classified in accordance with the people's way of thinking and understanding, and then to analyze the information from the face to understand people's emotions, involved in the field of biometric identification, pattern recognition, image processing, machine vision, motion tracking, physiology, psychology and other research challenging cross lesson is an important part of the affective computing, intelligent human-computer interaction, has a wide range of application prospects and potential market value. Firstly, summarized the research background and analysis at home and abroad have been proposed expression feature extraction methods and expression recognition. On this basis, the facial expression recognition method based on rough sets with mixed characteristics, the main contents are as follows: (1) the introduction of the rough set to facial expression recognition, an improved rough set attribute reduction algorithm to make up for the original heuristic attribute reduction algorithm only starting with the core attributes shortcomings, taking into account the importance of the property, but also consider the contribution of each attribute of information systems repeatability, can effectively select expression feature. (2) for the WPCA but stressed that a face image of one or two positions, the weighted characteristic shape of the area is not flexible enough shortcomings, an improved weighted principal component analysis algorithm (UWPCA) dispersed emphasis on changes in the expression an important contribution to the face in the eyes, eyebrows, mouth area, the facial expression feature more prominent; join the level of the vertical direction of the gradient scale factor, the weighted area two-way adjustable. (3) in order to extract more effective features, the integration based on local geometric features and global features mixed feature extraction method. AAM position the image sequences of the human face image feature points, select one of the 32 key points of the eyebrows, eyes, nose, mouth, cheek area, and then calculate the different distances of the different points of each image, and then rough set feature selection local geometric features extracted; UWPCA and rough set method of combining the overall characteristics of the expression image. The advantages and disadvantages of the two features complement each other, using the nuclear canonical correlation analysis to the integration of the two features, the fusion characteristics as the discrete HMM observation vector, to obtain better recognition rate. (4) the use of object-oriented design methods, design-based image sequence face expression recognition prototype system, the prototype system include: image pre-processing, expression feature extraction (AAM and UWPCA of operation), feature selection, emotional features of fusion and typical expressions identify functional modules.
|