The Research On Facial Expression Recognition Based On Collaborative Representation With Parameter Estimation

Along with the deep research in key techniques such as human-computer interaction,human emotion analysis and affective computing,facial expression recognition gets rapid development.Human facial expression is a nonverbal important way for people to communicate and understand emotion.As the facial expression recognition shows important value of theoretical research and practical application,it has gradually become a hotspot in the field of artificial intelligence and computer vision.In recent years,sparse representation based on compressive sensing has become popular and has shown an impressive performance in the field of image processing such as target tracking and face recognition.However,sparse representation based classification algorithm ignores the impact of the collaboration between classes for the classification.In contrast,the collaborative representation based classification algorithms not only take advantages of sparseness of the classification,but also take into account the collaboration between classes which enhance the classification.In view of the above problems,this thesis proposes a facial expression recognition algorithm based on collaborative representation with parameter estimation.The main contents of the thesis can be summarized as follows:(1)Based on l2 norm collaborative representation based classification model,this thesis estimates the norm approximation of the collaborative representation fidelity terms,and improves it by regularization factor.(2)Based on the classification model of collaborative representation,this thesis presents a regularized weighted collaborative representation classification model on the basis of maximum likelihood estimation,and describes its main workflow.By weighting iterative analysis of collaborative fidelity items,this model achieves the adaptive weight assignment of facial expression pixels,reduces the interference of boundary pixels.By maximizing the likelihood of collaborative fidelity items,this model minimizes the collaborative residuals and improves the effectiveness of facial expression recognition system effectively.(3)From the perspective of Bayesian estimation,this thesis presents a regularized weighted collaborative representation classification model on the basis of maximum a posteriori estimation.By introducing the priori factor and maximizing the posterior estimation of collaborative fidelity items,this model achieves the multi-angle and multi-level assessment of facial expression recognition system.By using the Newton iteration and estimating the parameters of the priori distribution,the model effectively simplifies the algorithm complexity of facial expression recognition system.The regularized weighted collaborative representation classification model and algorithm with maximum likelihood estimation can improve recognition accuracy and adaptability,and the regularized weighted collaborative representation classification model and algorithm with maximum a posteriori estimation can obtain the multi-angle and multi-level inspection.The above research supplies a kind of effective mechanism of the collaborative representation classification model,offers a high precision and robustness algorithm,which laid a solid foundation for the practical application of facial expression recognition system.