Face Recognition Based On Semi-supervised Sparisity Preserving Projections

With the development and application of computer technology,face recognition has been widely applied to every corner of human life.Despite the face recognition technology has got great development,how to process high-resolution face image rapidly is still a hot point.Principal Component Analysis(PCA),Linear Discriminant Analysis(LDA)and some other dimension reduction algorithms project high-dimensional face image dataset into the low-dimensional subspace.These methods obtain low-dimensional face feature.And it can reduce the time complexity of face recognition system and accelerate the speed of system.In order to further improve the accuracy of face recognition system,this paper uses Sparisity Preserving Projections(SPP),this method utilizes sparse representation to construct the sparse weight matrix among all face images.And SPP achieve dimension reduction through preserving the sparse weights between different face images.In the reality,face images with labels or without labels exist at the same time.This paper expands SPP into the framework of semi-supervised learning and realizes semi-supervised face recognition system.In order to deal with the problem that sparse weights matrix neglects the local structure of samples,this paper utilizes the similarity propagation which propagates sparse weights along the local structure.Semi-supervised Sparsity Preserving Projections(SSPP)fully constructs the correlations between face images.SSPP reduce the dimension of high-dimensional dataset through maintaining the correlations above in the original space.The low-dimensional space maintains the local topology structure of dataset and further improve the recognition accuracy.This paper introduces the whole face recognition system.A lot of experiments of this face recognition system verified that our proposed system can obtain a higher accuracy for face recognition.