The Research Of Facial Recognition Algorithm Based On Graph Regularized Dimensionality Reduction

Face recognition is one of the hot topics in the field of computer vision and pattern recognition.It becomes the most acceptable biometric technique for personal identification because of its reliability,non-touch and unduplicatedness.What's more,it has been attracted extensive attention from academia and business at home and abroad.So far,face recognition has already made great progress,but there are still huge challenges.As we all know,face is a natural structural object with a fairly complex detail change.The main challenges to identity such object are as follow.1)Variability:Face is non-rigid,its expression and posture will change with the changes of human emotions.2)Instability:It is inevitable to produce the shades for 3D facial images under different illumination angles and intensities,which seriously damages the facial images.3)Difference:Some adornments(such as sunglasses,masks or scarves,etc.)can form an occlusion on the face,which has an influence on the face recognition in real-world application.Combining with the above reasons,it is still a challenging subject that how to create a fast,efficient and stable face recognition system.This paper did some relative research on face recognition based on dimensionality reduction of graph regularized.The creative work of this paper includes:(1)Regularized Neighborhood Preserving Embedding Algorithm Based on QR DecompositionIn face recognition,facial images are high-dimensional multi-class and small size subjects.Therefore,the lack of adequate training samples is a thorny issue.Concerning the problem that estimation of the low-dimensional subspace data may have serious deviation under lacking of the training samples,a method of regularized neighborhood preserving embedding algorithm based on QR decomposition is proposed.Firstly,a local Laplace matrix is defined to preserve local structure of the original data.Secondly,the eigenspectrum of within-class scatter matrix is divided into three subspaces,the new eigenvector space obtained by inverse spectrum model defining weight function is used to preprocess the high-dimensional data.Finally,a neighborhood preserving adjacency matrix is defined,the projection matrix is obtained by QR decomposition and the nearest neighbor classifier is selected to face recognition.Compared with Regularized Generalized Discriminant Locality Preserving Projections(RGDLPP)algorithm,the recognition accuracy rate of the proposed method respectively increases 2%,1.5%,1.5%and 2%on ORL,Yale,FERET and PIE database.The experimental results illustrate that the proposed algorithm is easy to implement and has high recognition rate relatively under Small Sample Size.(2)Label Information-based Weighted Regularized Sparsity Preserving Embedding for Face RecognitionGraph embedding framework has become a popular method of dimensional reduction.However,the traditional graph construction heavily relies on the selection of parameters,resulting in unstable performance in real-world face recognition applications.To address this,a label information-based weighted regularized sparsity preserving embedding for face recognition is proposed in this paper.Different from the existing l1 graph,we adaptively construct both intrinsic graph and penalty graph with label information-based l1 graph under the graph embedding framework.In order to preserve local structure,Gaussian kernel distances between the samples are used as weight matrix to weight graph.In addition,the problem of irreversible matrix is alleviated by regularization instead of PCA that may lose some discriminative information.At last,an objective function combining globality and locality is created to reduce dimensionality.Meanwhile,Schmidt orthogonalization is used to obtain the orthogonal basis vectors.The experimental results on public face database illustrate that the proposed algorithm has high recognition rate.(3)Robust Face Recognition Based on Non-negative l1 Graph Regularized Low-rank RepresentationThe images may be affected by external environment(such as illuminations or occlusions)in the process of image acquisition so that both the training and test samples might be corrupted owing to illumination variations,noise and occlusions.To address this,a method of robust face recognition based on non-negative l1 graph regularized low-rank representation is proposed.Firstly,non-negative l1 graph regularization term is incorporated into the low-rank representation model,which recovers the clean samples and error samples from the training samples.Secondly,the sample dictionary and occlusion dictionary are built to decompose a test sample into non-occluded face image,occluded face image and error face image by sparse and collaborative representation model.Finally,the classification is carried out by calculating the residual between the recovered clean face image from test samples and the reconstruction test samples from sample dictionary and sparse coefficients.The experimental results on the Extended Yale B,CMU PIE and AR face databases demonstrate the effectiveness of the proposed algorithm and robustness to the noise.