A Facial Expression Recognition Method Based On The Gabor And2DPCA

Facial expression recognition is an important branch which is developed on the basisof face recognition. In daily life, expression expresses55%of the information, far higherthan7%which is communicated by language. Therefore, by the expression we can analyzea person’s heart activity more clearly and his views on a matter. Expression recognition isused in many fields, such as neural analysis, human-computer interaction, machine vision,behavioral science etc. Facial expression recognition has low recognition efficiencybecause of the expression feature dimension, so this paper proposes a facial expressionrecognition method based on the Gabor and2DPCA, using the nearest neighbor classifierand support vector machine to classify. Finally, this approach achieves94.29%recognitionrate.Firstly, the face images are preprocessed to obtain a normalized sample images. Imagepreprocessing is to split face region, normalize image size, adjust the image intensitychanges and enhance image contrast.Secondly, this paper uses the Gabor wavelet based on the physical characteristics toextract feature of facial expression. The texture features of expression image are extractedby Gabor filter banks in multi-scales and multi-orientations. With two-dimensionalprincipal component analysis (2DPCA) and the other two kinds of improved algorithm: lifttwo-dimensional principal component analysis(L2DPCA) and bilateral two-dimensionalprincipal component analysis(B2DPCA) respectively to further extract texture feature ofthe largest independent eigenvector. Then the extracted texture features are reduceddimensions by using PCA. So a facial image produces a row vector by feature extractionand dimension reduction, which is accorded to classify the expression image.Finally, this paper uses the nearest neighbor classifier and support vector machine forexpression classification.This experiment uses the Japanese Female Facial Expression and Facial ExpressionDatabase to experiment. Experimental result shows that the method not only effectivelyreduces the feature dimension of face image, but also retains the basic information ofcharacteristics, and this method has achieved to a high recognition rate. Compared with traditional2DPCA and L2DPCA, the B2DPCA remains less dimension of the featurematrix. The nearest neighbor classification is more suitable for facial expressionrecognition than support vector machine.