Research Of2D And3D Face Recognition Methods

With the development of society and the progress of computer technology, identityverification based on biological characters has attracted more and more researchers. Inall of the biological methods, face recognition has been the favor of researchers becauseof features: directness, friendliness and convenience. Due to traditional face recognitionsystems affected seriously by such as light, gesture and facial expression, designing asystem with strong robustness and high accuracy is still a problem to be resolved. In2Dface recognition, many great achievements have been acquired. However,2D faceimage can not offer enough information and the system is affected by posture and lighteffectively. With the development of3D technology,3D face has enough inherentinformation and appears to be less sensitive to light, gesture and so on.The thesis does some research separately on2D and3D face recognition anddesigns the face recognition for present deficiencies.For2D face recognition, gaussian radial basis function is usually applied as thekernel function of the kernel fisher discriminant analysis (KFD) in face recognitionapplication. However, the parameter of the kernel function has a great impact on theclassification. At present, the parameter is usually selected based on experience, and theprocess of KFD costs too much time for dealing with a large number of samples. Tosolve these problems, a face recognition method is presented with improved KFD andwavelet transform. It employs wavelet transform to compress the data of face image.And it applies PSO algorithm to automatically obtain the parameter to enhance theability of classification when KFD is employed to complete feature extraction. Finally,support vector machine is used for classification.For3D face recognition, a recognition algorithm based on the fusion of global andlocal information is proposed in this paper for improving the recognition accuracy interms of the scarcity of simplex feature information in3D face recognition. Firstly, wemake a surface fitting on the preprocessed3D cloudy points by using multilevelB-spline and acquire an accurate face surface fitting function. And then the controlpoints of the function are mapped into the range image, and the central profile and thehorizontal outline cross the nose top are extracted according to the surface function andphysiological characteristics of the face. Secondly,2D-PCA algorithm is applied on therange image for extracting global information, and the contours are matched using modified ICP algorithm as local information. Finally, the weighted sum method is usedto achieve information fusion at decision stage.