Research On Facial Expression Recognition Based On Feature Fusion

Natural harmony and intelligent human-machine interaction,the emotional communication between human and robot has always been an active research topic in artificial intelligence.As an important part of emotional computing and artificial intelligence,facial expression recognition has been aroused the attention of the majority of researchers.It is also a hot topic in image processing and computer vision.Therefore,the research on facial expression recognition has a great theoretical and practical significance.This thesis mainly focuses on the feature extraction and classification recognition in facial expression recognition,and aims to propose a more accurate and robust facial expression recognition algorithm.The main contents of this thesis are as follows:1.Curvelet transform is a very effective multi-scale analysis tool with bandpass and directional,which has been proposed to optimize the limitations of the image edge feature extracted from wavelet transform.On the basis of the study of Curvelet transformation for facial expression feature extraction,a Curvelet subband weighted feature fusion recognition algorithm is proposed,which can be used to improve the accuracy of classification rate by tuning the weights associating to the Low-frequency coefficients,Detail layer 1 coefficients,Detail layer 2 coefficients,High-frequency coefficients of Curvelet features.2.In order to obtain higher expression recognition rate and stronger robustness,a facial expression recognition algorithm based on Curvelet feature and sparse representation-based classifier(SRC)is proposed,which uses Curvelet feature,weighted fusion Curvelet feature and SRC to realize expression classification.The experimental results show that the our algorithm can represent test expression images with the aforementioned over-complete feature dictionary specifically,and it is insensitive to noise and error..3.It is worth nothing that sparse representation classify with minimum residual,a facial expression recognition algorithm based on feature fusion and sparse representation is proposed.Firstly,the algorithm framework fuse LBP feature and LPQ feature for better feature represent ability.Then,sparse representation residual are obtained from LBP/LPQ+SRC and Curvelet+SRC respectively,and construct residual rate according to residual information,the ratio of second minimum residual and minimum residual.Finally,if the classification labels of the two methods are not consistent,then it will be based on the maximum residual rate.Experiments show that the fusion algorithm of feature fusion and residual classification fusion can improve the recognition accuracy of facial expression effectively.