Research On Facial Expression Recognition Based On Gabor

Facial expression recognition is an important part of affective computing and intelligent Human-Computer Interactive,and is a cross-subject in the multiple fields of image processing,pattern recognition,machine learning,physiology and psychology. Because of the potential market value and wide application prospect,it is becoming the topics of most concern in academic community now.Facial expression recognition consists of such modules as image preprocessing, feature extraction and expression classification are studied.In this paper,several improved algorithms which refer to the critical problems existing in above three phases are presented. The major contributions of this paper are as follows:(1) An eye detection method using Zernike moments is applied to facial expression recognition system.Experimental results show that it can nicely process the problem of the expression image rotation.(2) A new eyes contour extraction method using PCA transform scattering projection (PCATSP) is proposed.Experimental results show that PCATSP is more robustly compared to deformable template for eye contour extraction in the shape of an average error of the parameters and optimizing the overall time.(3) A partitioning Boost algorithm used by feature selection is proposed, simultaneously,we design a partitioning Boost algorithm in combination with the Augmented Variance Ration.This paper adopts 5 scales and 8 orientations Gabor filters to extract the features of facial expression,and adopt the proposed method for feature selection to reduce the curse of dimensionality of features in facial expression recognition. Experimental results show that the proposed method not only acquires the vector MDGF, but achieves the purpose of dimensionality reduction. (4) We get the final recognition results by combining the priori knowledge of facial sample recognition results with its estimate value.This paper uses the one-against-one Support Vector Machine(SVM) to do the feature classification,and find the relationship between every scale,every orientation on vector MDGF and recognition results.Finally, we achieve the final recognition results by using the integration of the priori knowledge for sample recognition result and its estimation value,by deducing the computation cost and improving the classification precision.(5) Facial expression recognition system is developed.It consists of such modules as facial detection,eye detection,image preprocessing,character extraction and expression classification,which is platform for further research.