Design And Implementation Of Classification Algorithm Based On Low-Rank Constraint

Face recognition is the most popular biometric recognition technology because it is easy to collect and does not need to touch.In the process of face recognition,the design of classifier plays a key role,and feature representation also has an important impact on the performance of classifier.The performance of face recognition is easily affected by occlusion and erosion,which is the main challenge of face recognition.In order to improve the effectiveness of face recognition in practical application,many face recognition algorithms are innovated based on low-rank decomposition and low-rank representation,and it has been proved that it can effectively solve the problem of face image occlusion and corrosion.In this paper,based on the theory of low-rank matrix constraint,low-rank constraint is integrated into the traditional classification algorithm,and the face recognition classification algorithm is deeply studied The main work of this paper is as follows:1.Weighted non-negative sparse low-rank representation classification method with inconsistent structureDifferent classes of samples may have the same characteristics,which makes the existing low-rank representation classification methods have the problem of low discrimination and low interpretability.A weighted non-negative sparse low-rank representation classification method with inconsistent structure is proposed.The algorithm adds the constraint condition of inconsistent structure on the original basis,which can suppress the same features and retain the independent features,so that the obtained feature representation has better discrimination ability and interpretability.It is proved that the algorithm has superior recognition performance on ORL and AR databases2.Hypergraph laplacian regularized low-rank collaborative representation classification methodData samples are embedded in a low dimensional manifold in a high dimensional space and can be approximately represented by a low dimensional subspace.The traditional low-rank model ignores the nonlinear geometric structure between training samples,which makes the local and similar information lost in the learning process.In view of this problem,this paper proposes a hypergraph laplacian constrained low-rank collaborative representation classification,which can not only reveal the global low dimensional structure,but also capture the intrinsic nonlinear geometric structure of data.It is proved that the algorithm has superior recognition performance on CMU PIE and AR database.3.Subspace clustering method via learning an adaptive weighted low-rank graphGraph-based subspace clustering has attracted extensive attention in the field of computer vision in recent years because of its ability and efficiency of clustering data by low-rank representation of data.The two steps of representation and clustering are independent,so the overall optimal results cannot be guaranteed.Moreover,the parameters of the graph must be specified artificially in advance,so it is difficult to choose the best value.Therefore,a subspace clustering method via learning an adaptive weighted low-rank graph is proposed to learn affinity matrix and representation coefficient in a unified framework,which can effectively avoid the pre-calculated graph regularization term.At the same time,the weighted kernel norm is used to replace the traditional kernel norm,and different weights are assigned to different singular values,so the algorithm is more robust.It is proved that the algorithm has superior recognition performance on ORL and Extended Yale database.4.Face detection and location recognition systemA simple face recognition system is designed,which is mainly divided into the following functional parts.Firstly,the face image is preprocessed,such as similarity calculation and binarization.Then,the face region is detected to recognize face features,such as mouth,eyes and nose.Finally,the algorithm in this paper is used for classification and recognition.