Research Of Facial Expression Recognition Algorithm Based On Feature Fusion

With the rapid development and wide application of human-computer interaction technology, how to achieve the goal of interacting with a computer or robot harmoniously and smoothly has become the research emphasis for scholars at home and abroad. Facial expression recognition studied in this thesis is the key technology to solve the problem. Human emotional state changes are detected by the computer and robot precisely and quickly, thus ensuring human and computer interacts with each other in real-time effectively.This thesis focused on facial expression and made an intensive study of theories and methods concerning facial expression recognition. In order to solve the expression problem in face images, this thesis adopted the method of feature integration and combined the support vector machine based on the feature weighted, and finally realized the goal of classifying and recognizing facial expression and constructing a facial expression system.Firstly, according to the structure characteristics of different expression, active appearance model and weighted principal component analysis algorithm were used to extract expression feature. Meanwhile, in view of the shortcomings of the existing weighted principal component analysis algorithm, an improved weighted principal component analysis algorithm was used for expression feature extraction.Secondly, geometric characteristics extracted by active appearance model algorithm and statistical characteristics extracted by the improved weighted principal component analysis algorithm were selected by rough sets attribute reduction algorithm. Some effective information was kept and the redundant information was eliminated. An improved classic correlation analysis algorithm was adopted to realize the feature integration from the geometric characteristics and statistical characteristics retained by the aforementioned selection.Thirdly, the principle of traditional support vector machine and kernel function were studied. A combined kernel function based on the weighted feature was built on the basis of Gauss kernel function and polynomial kernel function. Precise and fast classification and recognition of human expression were achieved by applying it to support vector machine.Finally, in view of expression image sequences studied in this thesis, a facial expression recognition system was constructed, combining support vector machine face recognition framework based on feature integration and feature weighting. The system was not only used to validate the method adopted in this thesis, but also laid a solid foundation for further study.