Research On Linear And Nolinear Face Recognition On Subspace

With the progress of technology and development of the society, identity authentication in people's contact has become the core contents that building the credibility, face recognition technology emerges as the times require, feature extraction as the critical step has become a research hot spot. Because of the advantage of the better descriptive power, lower calculated amount , better stability and so on, subspace algorithms have become one of the most popular feature extraction algorithms. face recognition and feature extraction based on subspace are the focal study points in this paper, and developed algorithms have been proposed based on the traditional ones. Feature extraction' research and the solution to this kind of problems are very important to the development of pattern recognition.At the beginning, the domestic and foreign research status were introduced, and then this papaer discussed the importance of the feature extraction to the development of technology and society, mainly analysed some typical algorithms about face recognition and feature extraction, secondly, basic theories and algorithm's description are introduced, and this paper mainly researched on the exist problems of feature extraction, finally, three development face recognition algorithms based on the existing subspace theories were proposed.First, 2DGabor mean value algorithm is proposed based on the Gabor wavelet. This algorithm takes full account of the relationship between pixels, performs 2DGabor mean value transformation to every block of face image, it takes into accout the class information between image samples and improve the efficiency of edge feature extraction, so this algorithm has better recogniton capability.Second, the development bidirectional 2DPCA algorithm is proposed based on the bidirectional 2DPCA algorithm, this algorithm not only superimposed projection matrixs of rows and columns, but performed 2DPCA in the column direction on the eigenmatrix which have been performed 2DPCA in the row direction. This algorithm have better capability of compression computation.Third, Kernel Fisher Discriminant Analysis(KDA) combined with 2DGabor mean value algorithm is proposed based on the nonlinear discriminant analysis algorithms. KDA performs Fisher Discriminant Analysis(LDA) in the feature space which is bulit by the mapping of kernel function, this algorithm not only can describe the nonlinear face recognition well, but also has better separability between classes.