Research On The Application Of Extreme Learning Machine Data Dimensionality Reduction In Face Image Processing

In the application research of face image processing,the dimension reduction analysis and research of face data is one of its main contents.The dimensionality reduction methods of face data are mainly divided into linear dimensionality reduction and nonlinear dimensionality reduction.The global hypothesis of linear dimensionality reduction method cannot satisfy the geometric structure of face data,so nonlinear dimensionality reduction becomes a new research hotspot.Nonlinear dimensionality reduction methods mainly include kernel function dimensionality reduction and manifold learning,among which manifold learning is the most concerned.Many manifold learning algorithms can't solve out-of-sample problem because they have no explicit mapping function.In order to overcome this problem,we have proposed two new nonlinear explicit dimensionality reduction methods based on manifold learning and extreme learning machine.(1)Manifold preserving extreme learning machine(MP-ELM)is an unsupervised dimension reduction algorithm.MP-ELM algorithm uses the method of combining Euclidean distance and geodesic distance between sample points and neighboring sample points to reconstruct the neighborhood weight of sample points.MP-ELM not only considers the spatial structure of face data,but also considers the distance structure.It can fully mine the geometric structure of face data,so as to improve the dimensionality reduction performance of face data.Through the visualization experiment of artificial data and the dimension reduction experiment of ORL and Yale face data,it is proved that the MP-ELM algorithm proposed in this paper has better performance of dimension reduction.(2)In order to make up for the shortcomings of MP-ELM algorithm which cannot consider the category information,we introduce the category information to adjust the geodesic distance of sample points,and then uses rank order distance to calculate the distance between any two sample points,and puts forward SGRD-ELM algorithm based on discrimination information and neighbor sharing.SGRD-ELM not only considers the class information of the sample points,but also mines the neighborhood structure of the sample points by calculating the neighbor sharing information of the two sample points.The experiments on ORL and AR face data show that SGRD-ELM is better than MP-ELM.In order to show the effectiveness and applicability of SGRD-ELM algorithm in face image dimensionality reduction,we design a face image dimensionality reduction simulationrecognition system based on SGRD-ELM.First of all,we designed the MATLAB GUI platform according to the actual needs.Then the face recognition system is installed on the GUI platform.Finally,the ORL and a few self-built face images are recognized on the platform.Simulation results show that SGRD-ELM algorithm can effectively reduce the dimension of face image.