Facial Expression Recognition Algorithm Based On Spectral Features

In daily life and social intercourse,facial expressions containing a wealth of information can dynamically reflect one's emotion state and perceive state.In recent years,an increasing number of concerns have been focused on the analysis of facial expression,which has greatly enriched the computer vision,artificial intelligence and other fields.If computers could understand and deliver emotion just like humans, the relationship between human and machine would be changed radically,and then the computer could service us better.As a major component of the facial expression recognition system,feature extraction plays an important role on improving the recognition accuracy.In this paper, we focus on spectral feature analysis,and make some improvement and innovation based on that,and then imbed these algorithms in our system.The principal works are as follows:1.Improve the spectral feature analysis(SFA).Draw on the idea of spectral clustering,we make some improvements on the aspects of Laplacian matrix selected, Eigenvectors' direction and dimensions.The experiments on JAFFE database show that our method has improved on the computing speed and recognition accuracy to some extent,compared with traditional SFA and classical PCA.2.Propose a novel spectral feature based on the wavelet decomposition.Analyzing the relationship between facial regions related to the expression and all details of two-dimensional wavelet decomposition,we integrate the appropriate weighted components into a new image.And then take it as the input of our spectral feature method to improve the recognition accuracy further.3.Realize the above-mentioned methods in our system so that the performance and result of the algorithm can be showed intuitively.Experiments demonstrate that the improvement and introduced innovation are feasible and effective for expression recognition and have achieved satisfactory results.