Research On The Face Recognition Based On Spectral Regression

Face recognition technology has been one of the hot topics in some fields such as pattern recognition, image processing, computer vision, neural networks and cognitive science in recent years. It can be widely applied in records management systems, security verification systems, credit card verification, criminal identity recognition, monitoring in bank and customhouse, human-computer interaction, etc. Many face recognition techniques have been developed over the past few decades. Subspace learning based face recognition methods have attracted considerable interests in recent years, Including Principal Component Analysis(PCA), Linear Discriminant Analysis(LDA), and Locality Preserving Projection(LPP). However, a disadvantage of all these approaches is that the learned projective function are linear combinations of all the original features, thus it is often difficult to interpret the results. And the other disadvantage of some methods is that their computations involve eigen-decomposition of dense matrices. Details as follows:First, I introduce the graph embedding, and these are: Linearization, Kernelization, Tensorization. In graph embedding,each algorithm can be considered as the direct graph embedding or its linear/kernel/tensor extension of a specific intrinsic graph that describes certain desired statistical or geometric property of a data set, with constraints from scale normalization or a penalty graph that characterizes a statistical or geometric properties that should be avoided.And then the algorithm Spectral Regression (SR) is analyzed,and used in face recognition. Spectral methods have recently emerged as a powerful tool for dimensionality reduction and manifold learning. These methods use information contained in the eigenvectors of a data affinity matrix to reveal low dimensional structure in high dimensional data. SR casts the problem of learning an embedding function into a regression framework, which avoids eigen-decomposition of dense matrices.Then I rising a novel method which combined SR and 2DLPP, called it SR-2DLPP. Then obtain the solution using the regularization technique. The algorithm cast the problem of learning the projective functions into a regression framework,which avoid eigen-deposition of dense matrices. This algorithm can deal directly with image matrix, not only solved the singular value problem, and you don't need to use it to preprocessing. Also, with the regression based framework, different kinds of regularization can be naturally incorporated into the algorithm. Experimental results on classification and semi-supervised classification demonstrate the effectiveness and efficiency of the algorithm.