Research On Contourlet Transform And Locality Preserving Projection In Expression Recognition

A complete expression recognition system including capture the face expressionimage, pretreatment, face detection and localization, the face segmentation andnormalization, facial expression feature extraction, facial expression recognition. Thispaper studies the key issues about face expression feature extraction, feature selectionand expression classification, and put forward some improvement method and use thismethod imulation experiment.First, elaborated the research status of face expression and summarizes the mainmethods of face expression recognition at present, analysis the difficulties andcharacteristics of face expression recognition. Then, preprocess the expression images,convenient for the feature extraction.Next, introduce the theory of Contourlet transform. Contourlet transform is a newmultiscale geometric analysis method, it possess wavelet transform multiresolution andtime-frequency local characteristics, but also has very strong more directions andanisotropy. After Contourlet transform the image components low frequency which canreflect the general view images, and the high frequency sub-band direction can reflect theoutline of image and texture, and other details. And then processing and featureextraction. the expression images by using these characteristics of Contourlet transform.Because the dimension of expression features by Contourlet transform is bigger, so,this paper use locality preserving projection (LPP) algorithm reduced the expressionfeatures dimension. And introduces some theory of the manifold learning algorithm, atlast, improved the locality preserving projection (LPP) algorithm.Finally, according to these characteristics of the Contourlet transform andexpression images characteristics, this paper put forward segmented the face expressionimages for left eyes (including their eyebrows), right eyes (including their eyebrows) andmouth three parts. Processing the local expression images and the original images byContourlet transform, get the low frequency components and high frequency components.Combining the low frequency components of the local expression image with the high coefficients of the iginal image, then, using locality preserving projection (LPP)algorithm for expression feature extraction, finally using support vector machine (SVM)to classification.